Watermark embedder and reader

[0001] This application is a continuation-in-part of Ser. No. 09/761,349, filed Jan. 16, 2001 (now abandoned), which is a divisional of Ser. No. 09/127,502, filed Jul. 31, 1998 (now U.S. Pat. No. 6,345,104), which is a continuation-in-part of the following applications:

• • [0002] application Ser. No. 09/074,034, filed May 6, 1998 (now U.S. Pat. No. 6,449,377);
  • [0003] application Ser. No. 08/967,693, filed Nov. 12, 1997 (now U.S. Pat. No. 6,122,392), which is a continuation of application Ser. No. 08/614,521, filed Mar. 15, 1996 (now U.S. Pat. No. 5,745,604), which is a continuation of application Ser. No. 08/215,289, filed Mar. 17, 1994 (now abandoned); and
  • [0004] application Ser. No. 08/649,419, filed May 16, 1996 (now U.S. Pat. No. 5,862,260). Application Ser. No. 09/127,502 claims the benefit of provisional application 60/082,228, filed Apr. 16, 1998.

[0005] This patent application is also a continuation-in-part of Ser. No. 10/165,751, filed Jun. 6, 2002 (now U.S. Pat. No. 6,754,377), which is a continuation of Ser. No. 09/074,034, filed May 6, 1998 (now U.S. Pat. No. 6,449,377), which is a continuation-in-part of Ser. No. 08/438,159 (now U.S. Pat. No. 5,850,481), filed May 8, 1995, and claims priority to provisional application 60/082,228, filed Apr. 16, 1998.

[0006] This application is also a continuation of application Ser. No. 09/186,962, filed Nov. 5, 1998, (now U.S. Pat. No. 7,171,016), which is a continuation of application Ser. No. 08/649,419, filed May 16, 1996 (now U.S. Pat. No. 5,862,260), which is a continuation-in-part of the following applications: PCT/US96/06618, filed May 7, 1996; application Ser. No. 08/637,531, filed Apr. 25, 1996 (now U.S. Pat. No. 5,822,436); application Ser. No. 08/534,005, filed Sep. 25, 1995 (now U.S. Pat. No. 5,832,119); application Ser. No. 08/508,083, filed Jul. 27, 1995 (now U.S. Pat. No. 5,841,978); application Ser. No. 08/436,102, filed May 8, 1995 (now U.S. Pat. No. 5,748,783); and application Ser. No. 08/327,426, filed Oct. 21, 1994 (now U.S. Pat. No. 5,768,426). Application Ser. No. 08/327,426 is a continuation in part of application Ser. No. 08/215,289, filed Mar. 17, 1994 (now abandoned), which is a continuation in part of application Ser. No. 08/154,866, filed Nov. 18, 1993 (now abandoned). Application Ser. No. 08/637,531 also claims priority to application Ser. No. 08/512,993, filed Aug. 9, 1995 (now abandoned).

[0007] This application is also a continuation-in-part of Ser. No. 09/503,881, filed Feb. 14, 2000 (now U.S. Pat. No. 6,614,914), which is a continuation-in-part of Ser. No. 09/482,749, filed Jan. 13, 2000 (now abandoned).

[0008] This application is also a continuation-in-part of Ser. No. 09/612,177, filed Jul. 6, 2000 (now U.S. Pat. No. 6,681,029), which is a continuation of Ser. No. 08/746,613, filed Nov. 12, 1996 (now U.S. Pat. No. 6,122,403).

[0009] This application is also a continuation-in-part of Ser. No. 09/465,418, filed Dec. 16, 1999 (now abandoned), which claims priority to provisional application 60/112,955, filed Dec. 18, 1998.

[0010] This patent application is also a continuation-in-part of Ser. No. 09/498,223, filed Feb. 3, 2000 (now U.S. Pat. No. 6,574,350), which is continuation-in-part of Ser. No. 09/287,940, filed Apr. 7, 1999 (now U.S. Pat. No. 6,580,819), which claims priority to U.S. Provisional Application 60/082,228, filed Apr. 16, 1998. Application Ser. No. 09/498,223 is also a continuation of Ser. No. 09/433,104, filed Nov. 3, 1999 (now U.S. Pat. No. 6,636,615), which is continuation-in-part of application Ser. No. 09/234,780, filed Jan. 1, 1999 (now abandoned), which claims priority to U.S. Provisional Application 60/071,983, filed Jan. 20, 1998

[0011] The subject matter of this application is also related to that of the present assignee's other issued patents and applications: Ser. No. 09/616,462, U.S. Pat. Nos. 5,721,788 and 6,332,031. The above referenced patents and application are hereby incorporated by reference.

[0012] The file of this patent includes duplicate copies of a compact disc with a file entitled “Appendix B.txt”, created on Oct. 17, 2002, and having a size of 375,829 bytes (376,832 bytes on disc), which is hereby incorporated by reference. This source code originally formed part of the specification of related application Ser. Nos. 09/186,962 and 08/649,419, and is included to provide examples of digital watermarking embodiments.

[0013] The invention relates to digital watermarking of media content, such as images, audio and video.

[0014] Digital watermarking is a process for modifying media content to embed a machine-readable code into the data content. The data may be modified such that the embedded code is imperceptible or nearly imperceptible to the user, yet may be detected through an automated detection process. Most commonly, digital watermarking is applied to media such as images, audio signals, and video signals. However, it may also be applied to other types of data, including documents (e.g., through line, word or character shifting), software, multi-dimensional graphics models, and surface textures of objects.

[0015] Digital watermarking systems have two primary components: an embedding component that embeds the watermark in the media content, and a reading component that detects and reads the embedded watermark. The embedding component embeds a watermark pattern by altering data samples of the media content. The reading component analyzes content to detect whether a watermark pattern is present. In applications where the watermark encodes information, the reader extracts this information from the detected watermark.

[0016] One challenge to the developers of watermark embedding and reading systems is to ensure that the watermark is detectable even if the watermarked media content is transformed in some fashion. The watermark may be corrupted intentionally, so as to bypass its copy protection or anti-counterfeiting functions, or unintentionally through various transformations that result from routine manipulation of the content. In the case of watermarked images, such manipulation of the image may distort the watermark pattern embedded in the image.

[0017] This disclosure describes watermark structures, and related embedders, detectors, and readers for processing the watermark structures. In addition, it provides a variety of methods and applications associated with the watermark structures, embedders, detectors and readers. While adapted for images, the watermark system applies to other electronic and physical media. For example, it can be applied to electronic objects, including image, audio and video signals. It can be applied to mark blank paper, film and other substrates, and it can be applied by texturing object surfaces for a variety of applications, such as identification, authentication, etc. The detector and reader can operate on a signal captured from a physical object, even if that captured signal is distorted.

[0018] The watermark structure can have multiple components, each having different attributes. To name a few, these attributes include function, signal intensity, transform domain of watermark definition (e.g., temporal, spatial, frequency, etc.), location or orientation in host signal, redundancy, level of security (e.g., encrypted or scrambled). When describing a watermark signal in the context of this document, intensity refers to an embedding level while strength describes reading level (though the terms are sometimes used interchangeably). The components of the watermark structure may perform the same or different functions. For example, one component may carry a message, while another component may serve to identify the location or orientation of the watermark in a combined signal. Moreover, different messages may be encoded in different temporal or spatial portions of the host signal, such as different locations in an image or different time frames of audio or video.

[0019] Watermark components may have different signal intensities. For example, one component may carry a longer message, yet have smaller signal intensity than another component, or vice-versa. The embedder may adjust the signal intensity by encoding one component more redundantly than others, or by applying a different gain to the components. Additionally, watermark components may be defined in different transform domains. One may be defined in a frequency domain, while another may be defined in a spatial or temporal domain.

[0020] The watermark components may be located in different spatial or temporal locations in the host signal. In images, for example, different components may be located in different parts of the image. Each component may carry a different message or perform a different function. In audio or video, different components may be located in different time frames of the signal.

[0021] The watermark components may be defined, embedded and extracted in different domains. Examples of domains include spatial, temporal and frequency domains. A watermark may be defined in a domain by specifying how it alters the host signal in that domain to effect the encoding of the watermark component. A frequency domain component alters the signal in the frequency domain, while a spatial domain component alters the signal in the spatial domain. Of course, such alterations may have an impact that extends across many transform domains.

[0022] While described here as watermark components, one can also construe the components to be different watermarks. This enables the watermark technology described throughout this document to be used in applications using two or more watermarks. For example, some copy protection applications of the watermark structure may use two or more watermarks, each performing similar or different function. One mark may be more fragile than another, and thus, disappear when the combined signal is corrupted or transformed in some fashion. The presence or lack of a watermark or watermark component conveys information to the detector to initiate or prohibit some action, such as playback, copying or recording of the marked signal.

[0023] A watermark system may include an embedder, detector, and reader. The watermark embedder encodes a watermark signal in a host signal to create a combined signal. The detector looks for the watermark signal in a potentially corrupted version of the combined signal, and computes its orientation. Finally, a reader extracts a message in the watermark signal from the combined signal using the orientation to approximate the original state of the combined signal.

[0024] There are a variety of alternative embodiments of the embedder and detector. One embodiment of the embedder performs error correction coding of a binary message, and then combines the binary message with a carrier signal to create a component of a watermark signal. It then combines the watermark signal with a host signal. To facilitate detection, it may also add a detection component to form a composite watermark signal having a message and detection component. The message component includes known or signature bits to facilitate detection, and thus, serves a dual function of identifying the mark and conveying a message. The detection component is designed to identify the orientation of the watermark in the combined signal, but may carry an information signal as well. For example, the signal values at selected locations in the detection component can be altered to encode a message.

[0025] One embodiment of the detector estimates an initial orientation of a watermark signal in the multidimensional signal, and refines the initial orientation to compute a refined orientation. As part of the process of refining the orientation, this detector computes at least one orientation parameter that increases correlation between the watermark signal and the multidimensional signal when the watermark or multidimensional signal is adjusted with the refined orientation.

[0026] Another detector embodiment computes orientation parameter candidates of a watermark signal in different portions of the target signal, and compares the similarity of orientation parameter candidates from the different portions. Based on this comparison, it determines which candidates are more likely to correspond to a valid watermark signal.

[0027] Yet another detector embodiment estimates orientation of the watermark in a target signal suspected of having a watermark. The detector then uses the orientation to extract a measure of the watermark in the target. It uses the measure of the watermark to assess merits of the estimated orientation. In one implementation, the measure of the watermark is the extent to which message bits read from the target signal match with expected bits. Another measure is the extent to which values of the target signal are consistent with the watermark signal. The measure of the watermark signal provides information about the merits of a given orientation that can be used to find a better estimate of the orientation.

[0028] Further advantages and features of the invention will become apparent with reference to the following detailed description and accompanying drawings.

1.0 Introduction

[0061] A watermark can be viewed as an information signal that is embedded in a host signal, such as an image, audio, or some other media content. Watermarking systems based on the following detailed description may include the following components: 1) An embedder that inserts a watermark signal in the host signal to form a combined signal; 2) A detector that determines the presence and orientation of a watermark in a potentially corrupted version of the combined signal; and 3) A reader that extracts a watermark message from the combined signal. In some implementations, the detector and reader are combined.

[0062] The structure and complexity of the watermark signal can vary significantly, depending on the application. For example, the watermark may be comprised of one or more signal components, each defined in the same or different domains. Each component may perform one or more functions. Two primary functions include acting as an identifier to facilitate detection and acting as an information carrier to convey a message. In addition, components may be located in different spatial or temporal portions of the host signal, and may carry the same or different messages.

[0063] The host signal can vary as well. The host is typically some form of multi-dimensional media signal, such as an image, audio sequence or video sequence. In the digital domain, each of these media types is represented as a multi-dimensional array of discrete samples. For example, a color image has spatial dimensions (e.g., its horizontal and vertical components), and color space dimensions (e.g., YUV or RGB). Some signals, like video, have spatial and temporal dimensions. Depending on the needs of a particular application, the embedder may insert a watermark signal that exists in one or more of these dimensions.

[0064] In the design of the watermark and its components, developers are faced with several design issues such as: the extent to which the mark is impervious to jamming and manipulation (either intentional or unintentional); the extent of imperceptibility; the quantity of information content; the extent to which the mark facilitates detection and recovery, and the extent to which the information content can be recovered accurately.

[0065] For certain applications, such as copy protection or authentication, the watermark should be difficult to tamper with or remove by those seeking to circumvent it. To be robust, the watermark must withstand routine manipulation, such as data compression, copying, linear transformation, flipping, inversion, etc., and intentional manipulation intended to remove the mark or make it undetectable. Some applications require the watermark signal to remain robust through digital to analog conversion (e.g., printing an image or playing music), and analog to digital conversion (e.g., scanning the image or digitally sampling the music). In some cases, it is beneficial for the watermarking technique to withstand repeated watermarking.

[0066] A variety of signal processing techniques may be applied to address some or all of these design considerations. One such technique is referred to as spreading. Sometimes categorized as a spread spectrum technique, spreading is a way to distribute a message into a number of components (chips), which together make up the entire message. Spreading makes the mark more impervious to jamming and manipulation, and makes it less perceptible.

[0067] Another category of signal processing technique is error correction and detection coding. Error correction coding is useful to reconstruct the message accurately from the watermark signal. Error detection coding enables the decoder to determine when the extracted message has an error.

[0068] Another signal processing technique that is useful in watermark coding is called scattering. Scattering is a method of distributing the message or its components among an array of locations in a particular transform domain, such as a spatial domain or a spatial frequency domain. Like spreading, scattering makes the watermark less perceptible and more impervious to manipulation.

[0069] Yet another signal processing technique is gain control. Gain control is used to adjust the intensity of the watermark signal. The intensity of the signal impacts a number of aspects of watermark coding, including its perceptibility to the ordinary observer, and the ability to detect the mark and accurately recover the message from it.

[0070] Gain control can impact the various functions and components of the watermark differently. Thus, in some cases, it is useful to control the gain while taking into account its impact on the message and orientation functions of the watermark or its components. For example, in a watermark system described below, the embedder calculates a different gain for orientation and message components of an image watermark.

[0071] Another useful tool in watermark embedding and reading is perceptual analysis. Perceptual analysis refers generally to techniques for evaluating signal properties based on the extent to which those properties are (or are likely to be) perceptible to humans (e.g., listeners or viewers of the media content). A watermark embedder can take advantage of a Human Visual System (HVS) model to determine where to place a watermark and how to control the intensity of the watermark so that chances of accurately recovering the watermark are enhanced, resistance to tampering is increased, and perceptibility of the watermark is reduced. Such perceptual analysis can play an integral role in gain control because it helps indicate how the gain can be adjusted relative to the impact on the perceptibility of the mark. Perceptual analysis can also play an integral role in locating the watermark in a host signal. For example, one might design the embedder to hide a watermark in portions of a host signal that are more likely to mask the mark from human perception.

[0072] Various forms of statistical analyses may be performed on a signal to identify places to locate the watermark, and to identify places where to extract the watermark. For example, a statistical analysis can identify portions of a host image that have noise-like properties that are likely to make recovery of the watermark signal difficult. Similarly, statistical analyses may be used to characterize the host signal to determine where to locate the watermark.

[0073] Each of the techniques may be used alone, in various combinations, and in combination with other signal processing techniques.

[0074] In addition to selecting the appropriate signal processing techniques, the developer is faced with other design considerations. One consideration is the nature and format of the media content. In the case of digital images, for example, the image data is typically represented as an array of image samples. Color images are represented as an array of color vectors in a color space, such as RGB or YUV. The watermark may be embedded in one or more of the color components of an image. In some implementations, the embedder may transform the input image into a target color space, and then proceed with the embedding process in that color space.

2.0 Digital Watermark Embedder and Reader Overview

[0075] The following sections describe implementations of a watermark embedder and reader that operate on digital signals. The embedder encodes a message into a digital signal by modifying its sample values such that the message is imperceptible to the ordinary observer in output form. To extract the message, the reader captures a representation of the signal suspected of containing a watermark and then processes it to detect the watermark and decode the message.

[0076] FIG. 1 is a block diagram summarizing signal processing operations involved in embedding and reading a watermark. There are three primary inputs to the embedding process: the original, digitized signal 100, the message 102, and a series of control parameters 104. The control parameters may include one or more keys. One key or set of keys may be used to encrypt the message. Another key or set of keys may be used to control the generation of a watermark carrier signal or a mapping of information bits in the message to positions in a watermark information signal.

[0077] The carrier signal or mapping of the message to the host signal may be encrypted as well. Such encryption may increase security by varying the carrier or mapping for different components of the watermark or watermark message. Similarly, if the watermark or watermark message is redundantly encoded throughout the host signal, one or more encryption keys can be used to scramble the carrier or signal mapping for each instance of the redundantly encoded watermark. This use of encryption provides one way to vary the encoding of each instance of the redundantly encoded message in the host signal. Other parameters may include control bits added to the message, and watermark signal attributes (e.g., orientation or other detection patterns) used to assist in the detection of the watermark.

[0078] Apart from encrypting or scrambling the carrier and mapping information, the embedder may apply different, and possibly unique carrier or mapping for different components of a message, for different messages, or from different watermarks or watermark components to be embedded in the host signal. For example, one watermark may be encoded in a block of samples with one carrier, while another, possibly different watermark, is encoded in a different block with a different carrier. A similar approach is to use different mappings in different blocks of the host signal.

[0079] The watermark embedding process 106 converts the message to a watermark information signal. It then combines this signal with the input signal and possibly another signal (e.g., an orientation pattern) to create a watermarked signal 108. The process of combining the watermark with the input signal may be a linear or non-linear function. Examples of watermarking functions include: S*=S+gX; S*=S(1+gX); and S*=S e^gX; where S* is the watermarked signal vector, S is the input signal vector, and g is a function controlling watermark intensity. The watermark may be applied by modulating signal samples S in the spatial, temporal or some other transform domain.

[0080] To encode a message, the watermark encoder analyzes and selectively adjusts the host signal to give it attributes that correspond to the desired message symbol or symbols to be encoded. There are many signal attributes that may encode a message symbol, such as a positive or negative polarity of signal samples or a set of samples, a given parity (odd or even), a given difference value or polarity of the difference between signal samples (e.g., a difference between selected spatial intensity values or transform coefficients), a given distance value between watermarks, a given phase or phase offset between different watermark components, a modulation of the phase of the host signal, a modulation of frequency coefficients of the host signal, a given frequency pattern, a given quantizer (e.g., in Quantization Index Modulation) etc.

[0081] Some processes for combining the watermark with the input signal are termed non-linear, such as processes that employ dither modulation, modify least significant bits, or apply quantization index modulation. One type of non-linear modulation is where the embedder sets signal values so that they have some desired value or characteristic corresponding to a message symbol. For example, the embedder may designate that a portion of the host signal is to encode a given bit value. It then evaluates a signal value or set of values in that portion to determine whether they have the attribute corresponding to the message bit to be encoded. Some examples of attributes include a positive or negative polarity, a value that is odd or even, a checksum, etc. For example, a bit value may be encoded as a one or zero by quantizing the value of a selected sample to be even or odd. As another example, the embedder might compute a checksum or parity of an N bit pixel value or transform coefficient and then set the least significant bit to the value of the checksum or parity. Of course, if the signal already corresponds to the desired message bit value, it need not be altered. The same approach can be extended to a set of signal samples where some attribute of the set is adjusted as necessary to encode a desired message symbol. These techniques can be applied to signal samples in a transform domain (e.g., transform coefficients) or samples in the temporal or spatial domains.

[0082] Quantization index modulation techniques employ a set of quantizers. In these techniques, the message to be transmitted is used as an index for quantizer selection. In the decoding process, a distance metric is evaluated for all quantizers and the index with the smallest distance identifies the message value.

[0083] The watermark detector 110 operates on a digitized signal suspected of containing a watermark. As depicted generally in FIG. 1, the suspect signal may undergo various transformations 112, such as conversion to and from an analog domain, cropping, copying, editing, compression/decompression, transmission etc. Using parameters 114 from the embedder (e.g., orientation pattern, control bits, key(s)), it performs a series of correlation or other operations on the captured image to detect the presence of a watermark. If it finds a watermark, it determines its orientation within the suspect signal.

[0084] Using the orientation, if necessary, the reader 116 extracts the message. Some implementations do not perform correlation, but instead, use some other detection process or proceed directly to extract the watermark signal. For instance in some applications, a reader may be invoked one or more times at various temporal or spatial locations in an attempt to read the watermark, without a separate pre-processing stage to detect the watermark's orientation.

[0085] Some implementations require the original, un-watermarked signal to decode a watermark message, while others do not. In those approaches where the original signal is not necessary, the original un-watermarked signal can still be used to improve the accuracy of message recovery. For example, the original signal can be removed, leaving a residual signal from which the watermark message is recovered. If the decoder does not have the original signal, it can still attempt to remove portions of it (e.g., by filtering) that are expected not to contain the watermark signal.

[0086] Watermark decoder implementations use known relationships between a watermark signal and a message symbol to extract estimates of message symbol values from a signal suspected of containing a watermark. The decoder has knowledge of the properties of message symbols and how and where they are encoded into the host signal to encode a message. For example, it knows how message bit values of one and a zero are encoded and it knows where these message bits are originally encoded. Based on this information, it can look for the message properties in the watermarked signal. For example, it can test the watermarked signal to see if it has attributes of each message symbol (e.g., a one or zero) at a particular location and generate a probability measure as an indicator of the likelihood that a message symbol has been encoded. Knowing the approximate location of the watermark in the watermarked signal, the reader implementation may compare known message properties with the properties of the watermarked signal to estimate message values, even if the original signal is unavailable. Distortions to the watermarked signal and the host signal itself make the watermark difficult to recover, but accurate recovery of the message can be enhanced using a variety of techniques, such as error correction coding, watermark signal prediction, redundant message encoding, etc.

[0087] One way to recover a message value from a watermarked signal is to perform correlation between the known message property of each message symbol and the watermarked signal. If the amount of correlation exceeds a threshold, for example, then the watermarked signal may be assumed to contain the message symbol. The same process can be repeated for different symbols at various locations to extract a message. A symbol (e.g., a binary value of one or zero) or set of symbols may be encoded redundantly to enhance message recovery.

[0088] In some cases, it is useful to filter the watermarked signal to remove aspects of the signal that are unlikely to be helpful in recovering the message and/or are likely to interfere with the watermark message. For example, the decoder can filter out portions of the original signal and another watermark signal or signals. In addition, when the original signal is unavailable, the reader can estimate or predict the original signal based on properties of the watermarked signal. The original or predicted version of the original signal can then be used to recover an estimate of the watermark message. One way to use the predicted version to recover the watermark is to remove the predicted version before reading the desired watermark. Similarly, the decoder can predict and remove un-wanted watermarks or watermark components before reading the desired watermark in a signal having two or more watermarks.

2.1 Image Watermark Embedder

[0089] FIG. 2 is a block diagram illustrating an implementation of an exemplary embedder in more detail. The embedding process begins with the message 200. As noted above, the message is binary number suitable for conversion to a watermark signal. For additional security, the message, its carrier, and the mapping of the watermark to the host signal may be encrypted with an encryption key 202. In addition to the information conveyed in the message, the embedder may also add control bit values (“signature bits”) to the message to assist in verifying the accuracy of a read operation. These control bits, along with the bits representing the message, are input to an error correction coding process 204 designed to increase the likelihood that the message can be recovered accurately in the reader.

[0090] There are several alternative error correction coding schemes that may be employed. Some examples include BCH, convolution, Reed Solomon and turbo codes. These forms of error correction coding are sometimes used in communication applications where data is encoded in a carrier signal that transfers the encoded data from one place to another. In the digital watermarking application discussed here, the raw bit data is encoded in a fundamental carrier signal.

[0091] In addition to the error correction coding schemes mentioned above, the embedder and reader may also use a Cyclic Redundancy Check (CRC) to facilitate detection of errors in the decoded message data.

[0092] The error correction coding function 204 produces a string of bits, termed raw bits 206, that are embedded into a watermark information signal. Using a carrier signal 208 and an assignment map 210, the illustrated embedder encodes the raw bits in a watermark information signal 212, 214. In some applications, the embedder may encode a different message in different locations of the signal. The carrier signal may be a noise image. For each raw bit, the assignment map specifies the corresponding image sample or samples that will be modified to encode that bit.

[0093] The embedder depicted in FIG. 2 operates on blocks of image data (referred to as ‘tiles’) and replicates a watermark in each of these blocks. As such, the carrier signal and assignment map both correspond to an image block of a pre-determined size, namely, the size of the tile. To encode each bit, the embedder applies the assignment map to determine the corresponding image samples in the block to be modified to encode that bit. Using the map, it finds the corresponding image samples in the carrier signal. For each bit, the embedder computes the value of image samples in the watermark information signal as a function of the raw bit value and the value(s) of the corresponding samples in the carrier signal.

[0094] To illustrate the embedding process further, it is helpful to consider an example. First, consider the following background. Digital watermarking processes are sometimes described in terms of the transform domain in which the watermark signal is defined. The watermark may be defined in the spatial or temporal domain, or some other transform domain such as a wavelet transform, Discrete Cosine Transform (DCT), Discrete Fourier Transform (DFT), Hadamard transform, Hartley transform, Karhunen-Loeve transform (KLT) domain, etc.

[0095] Consider an example where the watermark is defined in a transform domain (e.g., a frequency domain such as DCT, wavelet or DFT). The embedder segments the image in the spatial domain into rectangular tiles and transforms the image samples in each tile into the transform domain. For example in the DCT domain, the embedder segments the image into N by N blocks and transforms each block into an N by N block of DCT coefficients. In this example, the assignment map specifies the corresponding sample location or locations in the frequency domain of the tile that correspond to a bit position in the raw bits. In the frequency domain, the carrier signal looks like a noise pattern. Each image sample in the frequency domain of the carrier signal is used together with a selected raw bit value to compute the value of the image sample at the location in the watermark information signal.

[0096] Now consider an example where the watermark is defined in the spatial domain. The embedder segments the image in the spatial domain into rectangular tiles of image samples (i.e. pixels). In this example, the assignment map specifies the corresponding sample location or locations in the tile that correspond to each bit position in the raw bits. In the spatial domain, the carrier signal looks like a noise pattern extending throughout the tile. Each image sample in the spatial domain of the carrier signal is used together with a selected raw bit value to compute the value of the image sample at the same location in the watermark information signal.

[0097] With this background, the embedder proceeds to encode each raw bit in the selected transform domain as follows. It uses the assignment map to look up the position of the corresponding image sample (or samples) in the carrier signal. The image sample value at that position in the carrier controls the value of the corresponding position in the watermark information signal. In particular, the carrier sample value indicates whether to invert the corresponding watermark sample value. The raw bit value is either a one or zero. Disregarding for a moment the impact of the carrier signal, the embedder adjusts the corresponding watermark sample upward to represent a one, or downward to represent a zero. Now, if the carrier signal indicates that the corresponding sample should be inverted, the embedder adjusts the watermark sample downward to represent a one, and upward to represent a zero. In this manner, the embedder computes the value of the watermark samples for a raw bit using the assignment map to find the spatial location of those samples within the block.

[0098] From this example, a number of points can be made. First, the embedder may perform a similar approach in any other transform domain. Second, for each raw bit, the corresponding watermark sample or samples are some function of the raw bit value and the carrier signal value. The specific mathematical relationship between the watermark sample, on one hand, and the raw bit value and carrier signal, on the other, may vary with the implementation. For example, the message may be convolved with the carrier, multiplied with the carrier, added to the carrier, or applied based on another non-linear function. Third, the carrier signal may remain constant for a particular application, or it may vary from one message to another. For example, a secret key may be used to generate the carrier signal. For each raw bit, the assignment map may define a pattern of watermark samples in the transform domain in which the watermark is defined. An assignment map that maps a raw bit to a sample location or set of locations (i.e. a map to locations in a frequency or spatial domain) is just one special case of an assignment map for a transform domain. Fourth, the assignment map may remain constant, or it may vary from one message to another. In addition, the carrier signal and map may vary depending on the nature of the underlying image. In sum, there many possible design choices within the implementation framework described above.

[0099] The embedder depicted in FIG. 2 combines another watermark component, shown as the detection watermark 216, with the watermark information signal to compute the final watermark signal. The detection watermark is specifically chosen to assist in identifying the watermark and computing its orientation in a detection operation.

[0100] FIG. 3 is a spatial frequency plot illustrating one quadrant of a detection watermark. The points in the plot represent impulse functions indicating signal content of the detection watermark signal. The pattern of impulse functions for the illustrated quadrant is replicated in all four quadrants. There are a number of properties of the detection pattern that impact its effectiveness for a particular application. The selection of these properties is highly dependent on the application. One property is the extent to which the pattern is symmetric about one or more axes. For example, if the detection pattern is symmetrical about the horizontal and vertical axes, it is referred to as being quad symmetric. If it is further symmetrical about diagonal axes at an angle of 45 degrees, it is referred to as being octally symmetric (repeated in a symmetric pattern 8 times about the origin). Such symmetry aids in identifying the watermark in an image, and aids in extracting the rotation angle. However, in the case of an octally symmetric pattern, the detector includes an additional step of testing which of the four quadrants the orientation angle falls into.

[0101] Another criterion is the position of the impulse functions and the frequency range that they reside in. Preferably, the impulse functions fall in a mid frequency range. If they are located in a low frequency range, they may be noticeable in the watermarked image. If they are located in the high frequency range, they are more difficult to recover. Also, they should be selected so that scaling, rotation, and other manipulations of the watermarked signal do not push the impulse functions outside the range of the detector. Finally, the impulse functions should preferably not fall on the vertical or horizontal axes, and each impulse function should have a unique horizontal and vertical location. While the example depicted in FIG. 3 shows that some of the impulse functions fall on the same horizontal axis, it is trivial to alter the position of the impulse functions such that each has a unique vertical or horizontal coordinate.

[0102] Returning to FIG. 2, the embedder makes a perceptual analysis 218 of the input image 220 to identify portions of the image that can withstand more watermark signal content without substantially impacting image fidelity. Generally, the perceptual analysis employs a HVS model to identify signal frequency bands and/or spatial areas to increase or decrease watermark signal intensity to make the watermark imperceptible to an ordinary observer. One type of model is to increase watermark intensity in frequency bands and spatial areas where there is more image activity. In these areas, the sample values are changing more than other areas and have more signal strength. The output of the perceptual analysis is a perceptual mask 222. The mask may be implemented as an array of functions, which selectively increase the signal strength of the watermark signal based on a HVS model analysis of the input image. The mask may selectively increase or decrease the signal strength of the watermark signal in areas of greater signal activity.

[0103] The embedder combines (224) the watermark information, the detection signal and the perceptual mask to yield the watermark signal 226. Finally, it combines (228) the input image 220 and the watermark signal 226 to create the watermarked image 230. In the frequency domain watermark example above, the embedder combines the transform domain coefficients in the watermark signal to the corresponding coefficients in the input image to create a frequency domain representation of the watermarked image. It then transforms the image into the spatial domain. As an alternative, the embedder may be designed to convert the watermark into the spatial domain, and then add it to the image.

[0104] In the spatial watermark example above, the embedder combines the image samples in the watermark signal to the corresponding samples in the input image to create the watermarked image 230.

[0105] The embedder may employ an invertible or non-invertible, and linear or non-linear function to combine the watermark signal and the input image (e.g., linear functions such as S*=S+gX; or S*=S(1+gX), convolution, quantization index modulation). The net effect is that some image samples in the input image are adjusted upward, while others are adjusted downward. The extent of the adjustment is greater in areas or subbands of the image having greater signal activity.

2.2. Overview of a Detector and Reader

[0106] FIG. 4 is a flow diagram illustrating an overview of a watermark detection process. This process analyzes image data 400 to search for an orientation pattern of a watermark in an image suspected of containing the watermark (the target image). First, the detector transforms the image data to another domain 402, namely the spatial frequency domain, and then performs a series of correlation or other detection operations 404. The correlation operations match the orientation pattern with the target image data to detect the presence of the watermark and its orientation parameters 406 (e.g., translation, scale, rotation, and differential scale relative to its original orientation). Finally, it re-orients the image data based on one or more of the orientation parameters 408.

[0107] If the orientation of the watermark is recovered, the reader extracts the watermark information signal from the image data (optionally by first re-orienting the data based on the orientation parameters). FIG. 5 is flow diagram illustrating a process of extracting a message from re-oriented image data 500. The reader scans the image samples (e.g., pixels or transform domain coefficients) of the re-oriented image (502), and uses known attributes of the watermark signal to estimate watermark signal values 504. Recall that in one example implementation described above, the embedder adjusted sample values (e.g., frequency coefficients, color values, etc.) up or down to embed a watermark information signal. The reader uses this attribute of the watermark information signal to estimate its value from the target image. Prior to making these estimates, the reader may filter the image to remove portions of the image signal that may interfere with the estimating process. For example, if the watermark signal is expected to reside in low or medium frequency bands, then high frequencies may be filtered out.

[0108] In addition, the reader may predict the value of the original un-watermarked image to enhance message recovery. One form of prediction uses temporal or spatial neighbors to estimate a sample value in the original image. In the frequency domain, frequency coefficients of the original signal can be predicted from neighboring frequency coefficients in the same frequency subband. In video applications for example, a frequency coefficient in a frame can be predicted from spatially neighboring coefficients within the same frame, or temporally neighboring coefficients in adjacent frames or fields. In the spatial domain, intensity values of a pixel can be estimated from intensity values of neighboring pixels. Having predicted the value of a signal in the original, un-watermarked image, the reader then estimates the watermark signal by calculating an inverse of the watermarking function used to combine the watermark signal with the original signal.

[0109] For such watermark signal estimates, the reader uses the assignment map to find the corresponding raw bit position and image sample in the carrier signal (506). The value of the raw bit is a function of the watermark signal estimate, and the carrier signal at the corresponding location in the carrier. To estimate the raw bit value, the reader solves for its value based on the carrier signal and the watermark signal estimate. As reflected generally in FIG. 5 (508), the result of this computation represents only one estimate to be analyzed along with other estimates impacting the value of the corresponding raw bit. Some estimates may indicate that the raw bit is likely to be a one, while others may indicate that it is a zero. After the reader completes its scan, it compiles the estimates for each bit position in the raw bit string, and makes a determination of the value of each bit at that position (510). Finally, it performs the inverse of the error correction coding scheme to construct the message (512). In some implementations, probablistic models may be employed to determine the likelihood that a particular pattern of raw bits is just a random occurrence rather than a watermark.

2.2.1 Example Illustrating Detector Process

[0110] FIG. 6 is a diagram depicting an example of a watermark detection process. The detector segments the target image into blocks (e.g., 600, 602) and then performs a 2-dimensional fast fourier transform (2D FFT) on several blocks. This process yields 2D transforms of the magnitudes of the image contents of the blocks in the spatial frequency domain as depicted in the plot 604 shown in FIG. 6.

[0111] Next, the detector process performs a log polar remapping of each transformed block. The detector may add some of the blocks together to increase the watermark signal to noise ratio. The type of remapping in this implementation is referred to as a Fourier Mellin transform. The Fourier Mellin transform is a geometric transform that warps the image data from a frequency domain to a log polar coordinate system. As depicted in the plot 606 shown in FIG. 6, this transform sweeps through the transformed image data along a line at angle θ, mapping the data to a log polar coordinate system shown in the next plot 608. The log polar coordinate system has a rotation axis, representing the angle θ, and a scale axis. Inspecting the transformed data at this stage, one can see the orientation pattern of the watermark begin to be distinguishable from the noise component (i.e., the image signal).

[0112] Next, the detector performs a correlation 610 between the transformed image block and the transformed orientation pattern 612. At a high level, the correlation process slides the orientation pattern over the transformed image (in a selected transform domain, such as a spatial frequency domain) and measures the correlation at an array of discrete positions. Each such position has a corresponding scale and rotation parameter associated with it. Ideally, there is a position that clearly has the highest correlation relative to all of the others. In practice, there may be several candidates with a promising measure of correlation. As explained further below, these candidates may be subjected to one or more additional correlation stages to select the one that provides the best match.

[0113] There are a variety of ways to implement the correlation process. Any number of generalized matching filters may be implemented for this purpose. One such filter performs an FFT on the target and the orientation pattern, and multiplies the resulting arrays together to yield a multiplied FFT. Finally, it performs an inverse FFT on the multiplied FFT to return the data into its original log-polar domain. The position or positions within this resulting array with the highest magnitude represent the candidates with the highest correlation.

[0114] When there are several viable candidates, the detector can select a set of the top candidates and apply an additional correlation stage. Each candidate has a corresponding rotation and scale parameter. The correlation stage rotates and scales the FFT of the orientation pattern and performs a matching operation with the rotated and scaled pattern on the FFT of the target image. The matching operation multiplies the values of the transformed pattern with sample values at corresponding positions in the target image and accumulates the result to yield a measure of the correlation. The detector repeats this process for each of the candidates and picks the one with the highest measure of correlation. As shown in FIG. 6, the rotation and scale parameters (614) of the selected candidate are then used to find additional parameters that describe the orientation of the watermark in the target image.

[0115] The detector applies the scale and rotation to the target data block 616 and then performs another correlation process between the orientation pattern 618 and the scaled and rotated data block 616. The correlation process 620 is a generalized matching filter operation. It provides a measure of correlation for an array of positions that each has an associated translation parameter (e.g., an x, y position). Again, the detector may repeat the process of identifying promising candidates (i.e. those that reflect better correlation relative to others) and using those in an additional search for a parameter or set of orientation parameters that provide a better measure of correlation.

[0116] At this point, the detector has recovered the following orientation parameters: rotation, scale and translation. For many applications, these parameters may be sufficient to enable accurate reading of the watermark. In the read operation, the reader applies the orientation parameters to re-orient the target image and then proceeds to extract the watermark signal.

[0117] In some applications, the watermarked image may be stretched more in one spatial dimension than another. This type of distortion is sometimes referred to as differential scale or shear. Consider that the original image blocks are square. As a result of differential scale, each square may be warped into a parallelogram with unequal sides. Differential scale parameters define the nature and extent of this stretching.

[0118] There are several alternative ways to recover the differential scale parameters. One general class of techniques is to use the known parameters (e.g., the computed scale, rotation, and translation) as a starting point to find the differential scale parameters. Assuming the known parameters to be valid, this approach warps either the orientation pattern or the target image with selected amounts of differential scale and picks the differential scale parameters that yield the best correlation.

[0119] Another approach to determination of differential scale is set forth in application Ser. No. 09/452,022, filed Nov. 30, 1999 (now U.S. Pat. No. 6,959,098), which is hereby incorporated by reference.

2.2.2 Example Illustrating Reader Process

[0120] FIG. 7 is a diagram illustrating a re-oriented image 700 superimposed onto the original watermarked image 702. The difference in orientation and scale shows how the image was transformed and edited after the embedding process. The original watermarked image is sub-divided into tiles (e.g., pixel blocks 704, 706, etc.). When superimposed on the coordinate system of the original image 702 shown in FIG. 7, the target image blocks typically do not match the orientation of the original blocks.

[0121] The reader scans samples of the re-oriented image data, estimating the watermark information signal. It estimates the watermark information signal, in part, by predicting original sample values of the un-watermarked image. The reader then uses an inverted form of the watermarking function to estimate the watermark information signal from the watermarked signal and the predicted signal. This inverted watermarking function expresses the estimate of the watermark signal as a function of the predicted signal and the watermarked signal. Having an estimate of the watermark signal, it then uses the known relationship among the carrier signal, the watermark signal, and the raw bit to compute an estimate of the raw bit. Recall that samples in the watermark information signal are a function of the carrier signal and the raw bit value. Thus, the reader may invert this function to solve for an estimate of the raw bit value.

[0122] Recall that the embedder implementation discussed in connection with FIG. 2 redundantly encodes the watermark information signal in blocks of the input signal. Each raw bit may map to several samples within a block. In addition, the embedder repeats a mapping process for each of the blocks. As such, the reader generates several estimates of the raw bit value as it scans the watermarked image.

[0123] The information encoded in the raw bit string can be used to increase the accuracy of read operations. For instance, some of the raw bits act as signature bits that perform a validity checking function. Unlike unknown message bits, the reader knows the expected values of these signature bits. The reader can assess the validity of a read operation based on the extent to which the extracted signature bit values match the expected signature bit values. The estimates for a given raw bit value can then be given a higher weight depending on whether they are derived from a tile with a greater measure of validity.

3.0 Embedder Implementation:

[0124] The following sections describe an implementation of the digital image watermark embedder depicted in FIG. 8. The embedder inserts two watermark components into the host image: a message component and a detection component (called the orientation pattern). The message component is defined in a spatial domain or other transform domain, while the orientation pattern is defined in a frequency domain. As explained later, the message component serves a dual function of conveying a message and helping to identify the watermark location in the image.

[0125] The embedder inserts the watermark message and orientation pattern in blocks of a selected color plane or planes (e.g., luminance or chrominance plane) of the host image. The message payload varies from one application to another, and can range from a single bit to the number of image samples in the domain in which it is embedded. The blocks may be blocks of samples in a spatial domain or some other transform domain.

3.1 Encoding the Message

[0126] The embedder converts binary message bits into a series of binary raw bits that it hides in the host image. As part of this process, a message encoder 800 appends certain known bits to the message bits 802. It performs an error detection process (e.g., parity, Cyclic Redundancy Check (CRC), etc.) to generate error detection bits and adds the error detection bits to the message. An error correction coding operation then generates raw bits from the combined known and message bit string.

[0127] For the error correction operation, the embedder may employ any of a variety of error correction codes such as Reed Solomon, BCH, convolution or turbo codes. The encoder may perform an M-ary modulation process on the message bits that maps groups of message bits to a message signal based on an M-ary symbol alphabet.

[0128] In one application of the embedder, the component of the message representing the known bits is encoded more redundantly than the other message bits. This is an example of a shorter message component having greater signal strength than a longer, weaker message component. The embedder gives priority to the known bits in this scheme because the reader uses them to verify that it has found the watermark in a potentially corrupted image, rather than a signal masquerading as the watermark.

3.2 Spread Spectrum Modulation

[0129] The embedder uses spread spectrum modulation as part of the process of creating a watermark signal from the raw bits. A spread spectrum modulator 804 spreads each raw bit into a number of “chips.” The embedder generates a pseudo random number that acts as the carrier signal of the message. To spread each raw bit, the modulator performs an exclusive OR (XOR) operation between the raw bit and each bit of a pseudo random binary number of a pre-determined length. The length of the pseudo random number depends, in part, on the size of the message and the image. Preferably, the pseudo random number should contain roughly the same number of zeros and ones, so that the net effect of the raw bit on the host image block is zero. If a bit value in the pseudo random number is a one, the value of the raw bit is inverted. Conversely, if the bit value is a zero, then the value of the raw bit remains the same.

[0130] The length of the pseudorandom number may vary from one message bit or symbol to another. By varying the length of the number, some message bits can be spread more than others.

3.3 Scattering the Watermark Message

[0131] The embedder scatters each of the chips corresponding to a raw bit throughout an image block. An assignment map 806 assigns locations in the block to the chips of each raw bit. Each raw bit is spread over several chips. As noted above, an image block may represent a block of transform domain coefficients or samples in a spatial domain. The assignment map may be used to encode some message bits or symbols (e.g., groups of bits) more redundantly than others by mapping selected bits to more locations in the host signal than other message bits. In addition, it may be used to map different messages, or different components of the same message, to different locations in the host signal.

[0132] FIG. 9 depicts an example of the assignment map. Each of the blocks in FIG. 9 correspond to an image block and depict a pattern of chips corresponding to a single raw bit. FIG. 9 depicts a total of 32 example blocks. The pattern within a block is represented as black dots on a white background. Each of the patterns is mutually exclusive such that each raw bit maps to a pattern of unique locations relative to the patterns of every other raw bit. Though not a requirement, the combined patterns, when overlapped, cover every location within the image block.

3.4 Gain Control and Perceptual Analysis

[0133] To insert the information carried in a chip to the host image, the embedder alters the corresponding sample value in the host image. In particular, for a chip having a value of one, it adds to the corresponding sample value, and for a chip having a value of zero, it subtracts from the corresponding sample value. A gain controller in the embedder adjusts the extent to which each chip adds or subtracts from the corresponding sample value.

[0134] The gain controller takes into account the orientation pattern when determining the gain. It applies a different gain to the orientation pattern than to the message component of the watermark. After applying the gain, the embedder combines the orientation pattern and message components together to form the composite watermark signal, and combines the composite watermark with the image block. One way to combine these signal components is to add them, but other linear or non-linear functions may be used as well.

[0135] The orientation pattern is comprised of a pattern of quad symmetric impulse functions in the spatial frequency domain. In the spatial domain, these impulse functions look like cosine waves. An example of the orientation pattern is depicted in FIGS. 10 and 11. FIG. 10 shows the impulse functions as points in the spatial frequency domain, while FIG. 11 shows the orientation pattern in the spatial domain. Before adding the orientation pattern component to the message component, the embedder may transform the watermark components to a common domain. For example, if the message component is in a spatial domain and the orientation component is in a frequency domain, the embedder transforms the orientation component to a common spatial domain before combining them together.

[0136] FIG. 8 depicts the gain controller used in the embedder. Note that the gain controller operates on the blocks of image samples 808, the message watermark signal, and a global gain input 810, which may be specified by the user. A perceptual analyzer component 812 of the gain controller performs a perceptual analysis on the block to identify samples that can tolerate a stronger watermark signal without substantially impacting visibility. In places where the naked eye is less likely to notice the watermark, the perceptual analyzer increases the strength of the watermark. Conversely, it decreases the watermark strength where the eye is more likely to notice the watermark.

[0137] The perceptual analyzer shown in FIG. 8 performs a series of filtering operations on the image block to compute an array of gain values. There are a variety of filters suitable for this task. These filters include an edge detector filter that identifies edges of objects in the image, a non-linear filter to map gain values into a desired range, and averaging or median filters to smooth the gain values. Each of these filters may be implemented as a series of one-dimensional filters (one operating on rows and the other on columns) or two-dimensional filters. The size of the filters (i.e. the number of samples processed to compute a value for a given location) may vary (e.g., 3 by 3, 5 by 5, etc.). The shape of the filters may vary as well (e.g., square, cross-shaped, etc.). The perceptual analyzer process produces a detailed gain multiplier. The multiplier is a vector with elements corresponding to samples in a block.

[0138] Another component 818 of the gain controller computes an asymmetric gain based on the output of the image sample values and message watermark signal.

[0139] This component analyzes the samples of the block to determine whether they are consistent with the message signal. The embedder reduces the gain for samples whose values relative to neighboring values are consistent with the message signal.

[0140] The embedder applies the asymmetric gain to increase the chances of an accurate read in the watermark reader. To understand the effect of the asymmetric gain, it is helpful to explain the operation of the reader. The reader extracts the watermark message signal from the watermarked signal using a predicted version of the original signal. It estimates the watermark message signal value based on values of the predicted signal and the watermarked signal at locations of the watermarked signal suspected of containing a watermark signal. There are several ways to predict the original signal. One way is to compute a local average of samples around the sample of interest. The average may be computed by taking the average of vertically adjacent samples, horizontally adjacent samples, an average of samples in a cross-shaped filter (both vertical and horizontal neighbors, an average of samples in a square-shaped filter, etc. The estimate may be computed one time based on a single predicted value from one of these averaging computations. Alternatively, several estimates may be computed based on two or more of these averaging computations (e.g., one estimate for vertically adjacent samples and another for horizontally adjacent samples). In the latter case, the reader may keep estimates if they satisfy a similarity metric. In other words, the estimates are deemed valid if they within a predetermined value or have the same polarity.

[0141] Knowing this behavior of the reader, the embedder computes the asymmetric gain as follows. For samples that have values relative to their neighbors that are consistent with the watermark signal, the embedder reduces the asymmetric gain. Conversely, for samples that are inconsistent with the watermark signal, the embedder increases the asymmetric gain. For example, if the chip value is a one, then the sample is consistent with the watermark signal if its value is greater than its neighbors. Alternatively, if the chip value is a zero, then the sample is consistent with the watermark signal if its value is less than its neighbors.

[0142] Another component 820 of the gain controller computes a differential gain, which represents an adjustment in the message vs. orientation pattern gains. As the global gain increases, the embedder emphasizes the message gain over the orientation pattern gain by adjusting the global gain by an adjustment factor. The inputs to this process 820 include the global gain 810 and a message differential gain 822. When the global gain is below a lower threshold, the adjustment factor is one. When the global gain is above an upper threshold, the adjustment factor is set to an upper limit greater than one. For global gains falling within the two thresholds, the adjustment factor increases linearly between one and the upper limit. The message differential gain is the product of the adjustment factor and the global gain.

[0143] At this point, there are four sources of gain: the detailed gain, the global gain, the asymmetric gain, and the message dependent gain. The embedder applies the first two gain quantities to both the message and orientation watermark signals. It only applies the latter two to the message watermark signal. FIG. 8 depicts how the embedder applies the gain to the two watermark components. First, it multiplies the detailed gain with the global gain to compute the orientation pattern gain. It then multiplies the orientation pattern gain with the adjusted message differential gain and asymmetric gain to form the composite message gain.

[0144] Finally, the embedder forms the composite watermark signal. It multiplies the composite message gain with the message signal, and multiplies the orientation pattern gain with the orientation pattern signal. It then combines the result in a common transform domain to get the composite watermark. The embedder applies a watermarking function to combine the composite watermark to the block to create a watermarked image block. The message and orientation components of the watermark may be combined by mapping the message bits to samples of the orientation signal, and modulating the samples of the orientation signal to encode the message.

[0145] The embedder computes the watermark message signal by converting the output of the assignment map 806 to delta values, indicating the extent to which the watermark signal changes the host signal. As noted above, a chip value of one corresponds to an upward adjustment of the corresponding sample, while a chip value of zero corresponds to a downward adjustment. The embedder specifies the specific amount of adjustment by assigning a delta value to each of the watermark message samples (830).

4.0 Detector Implementation

[0146] FIG. 12 illustrates an overview of a watermark detector that detects the presence of a detection watermark in a host image and its orientation. Using the orientation pattern and the known bits inserted in the watermark message, the detector determines whether a potentially corrupted image contains a watermark, and if so, its orientation in the image.

[0147] Recall that the composite watermark is replicated in blocks of the original image. After an embedder places the watermark in the original digital image, the watermarked image is likely to undergo several transformations, either from routine processing or from intentional tampering. Some of these transformations include: compression, decompression, color space conversion, digital to analog conversion, printing, scanning, analog to digital conversion, scaling, rotation, inversion, flipping differential scale, and lens distortion. In addition to these transformations, various noise sources can corrupt the watermark signal, such as fixed pattern noise, thermal noise, etc.

[0148] When building a detector implementation for a particular application, the developer may implement counter-measures to mitigate the impact of the types of transformations, distortions and noise expected for that application. Some applications may require more counter-measures than others. The detector described below is designed to recover a watermark from a watermarked image after the image has been printed, and scanned. The following sections describe the counter-measures to mitigate the impact of various forms of corruption. The developer can select from among these counter-measures when implementing a detector for a particular application.

[0149] For some applications, the detector will operate in a system that provides multiple image frames of a watermarked object. One typical example of such a system is a computer equipped with a digital camera. In such a configuration, the digital camera can capture a temporal sequence of images as the user or some device presents the watermarked image to the camera.

[0150] As shown in FIG. 12, the principal components of the detector are: 1) pre-processor 900; 2) rotation and scale estimator 902; 3) orientation parameter refiner 904; 4) translation estimator 906; 5) translation refiner 908; and reader 910.

[0151] The preprocessor 900 takes one or more frames of image data 912 and produces a set of image blocks 914 prepared for further analysis. The rotation-scale estimator 902 computes rotation-scale vectors 916 that estimate the orientation of the orientation signal in the image blocks. The parameter refiner 904 collects additional evidence of the orientation signal and further refines the rotation scale vector candidates by estimating differential scale parameters. The result of this refining stage is a set of 4D vectors candidates 918 (rotation, scale, and two differential scale parameters). The translation estimator 906 uses the 4D vector candidates to re-orient image blocks with promising evidence of the orientation signal. It then finds estimates of translation parameters 920. The translation refiner 908 invokes the reader 910 to assess the merits of an orientation vector. When invoked by the detector, the reader uses the orientation vector to approximate the original orientation of the host image and then extracts values for the known bits in the watermark message. The detector uses this information to assess the merits of and refine orientation vector candidates.

[0152] By comparing the extracted values of the known bits with the expected values, the reader provides a figure of merit for an orientation vector candidate. The translation refiner then picks a 6D vector, including rotation, scale, differential scale and translation, that appears likely produce a valid read of the watermark message 922. The following sections describe implementations of these components in more detail.

4.1 Detector Pre-Processing

[0153] FIG. 13 is a flow diagram illustrating preprocessing operations in the detector shown in FIG. 12. The detector performs a series of pre-processing operations on the native image 930 to prepare the image data for further analysis. It begins by filling memory with one or more frames of native image data (932), and selecting sets of pixel blocks 934 from the native image data for further analysis (936). While the detector can detect a watermark using a single image frame, it also has support for detecting the watermark using additional image frames. As explained below, the use of multiple frames has the potential for increasing the chances of an accurate detection and read.

[0154] In applications where a camera captures an input image of a watermarked object, the detector may be optimized to address problems resulting from movement of the object. Typical PC cameras, for example, are capable of capturing images at a rate of at least 10 frames a second. A frustrated user might attempt to move the object in an attempt to improve detection. Rather than improving the chances of detection, the movement of the object changes the orientation of the watermark from one frame to the next, potentially making the watermark more difficult to detect. One way to address this problem is to buffer one or more frames, and then screen the frame or frames to determine if they are likely to contain a valid watermark signal. If such screening indicates that a frame is not likely to contain a valid signal, the detector can discard it and proceed to the next frame in the buffer, or buffer a new frame. Another enhancement is to isolate portions of a frame that are most likely to have a valid watermark signal, and then perform more detailed detection of the isolated portions.

[0155] After loading the image into the memory, the detector selects image blocks 934 for further analysis. It is not necessary to load or examine each block in a frame because it is possible to extract the watermark using only a portion of an image. The detector looks at only a subset of the samples in an image, and preferably analyzes samples that are more likely to have a recoverable watermark signal.

[0156] The detector identifies portions of the image that are likely to have the highest watermark signal to noise ratio. It then attempts to detect the watermark signal in the identified portions. In the context of watermark detection, the host image is considered to be a source of noise along with conventional noise sources. While it is typically not practical to compute the signal to noise ratio, the detector can evaluate attributes of the signal that are likely to evince a promising watermark signal to noise ratio. These properties include the signal activity (as measured by sample variance, for example), and a measure of the edges (abrupt changes in image sample values) in an image block. Preferably, the signal activity of a candidate block should fall within an acceptable range, and the block should not have a high concentration of strong edges. One way to quantify the edges in the block is to use an edge detection filter (e.g., a LaPlacian, Sobel, etc.).

[0157] In one implementation, the detector divides the input image into blocks, and analyzes each block based on pre-determined metrics. It then ranks the blocks according to these metrics. The detector then operates on the blocks in the order of the ranking. The metrics include sample variance in a candidate block and a measure of the edges in the block. The detector combines these metrics for each candidate block to compute a rank representing the probability that it contains a recoverable watermark signal.

[0158] In another implementation, the detector selects a pattern of blocks and evaluates each one to try to make the most accurate read from the available data. In either implementation, the block pattern and size may vary. This particular implementation selects a pattern of overlapping blocks (e.g., a row of horizontally aligned, overlapping blocks). One optimization of this approach is to adaptively select a block pattern that increases the signal to noise ratio of the watermark signal. While shown as one of the initial operations in the preparation, the selection of blocks can be postponed until later in the pre-processing stage.

[0159] Next, the detector performs a color space conversion on native image data to compute an array of image samples in a selected color space for each block (936). In the following description, the color space is luminance, but the watermark may be encoded in one or more different color spaces. The objective is to get a block of image samples with lowest noise practical for the application. While the implementation currently performs a row by row conversion of the native image data into 8 bit integer luminance values, it may be preferable to convert to floating-point values for some applications. One optimization is to select a luminance converter that is adapted for the sensor used to capture the digital input image. For example, one might experimentally derive the lowest noise luminance conversion for commercially available sensors, e.g., CCD cameras or scanners, CMOS cameras, etc. Then, the detector could be programmed to select either a default luminance converter, or one tuned to a specific type of sensor.

[0160] At one or more stages of the detector, it may be useful to perform operations to mitigate the impact of noise and distortion. In the pre-processing phase, for example, it may be useful to evaluate fixed pattern noise and mitigate its effect (938). The detector may look for fixed pattern noise in the native input data or the luminance data, and then mitigate it.

[0161] One way to mitigate certain types of noise is to combine data from different blocks in the same frame, or corresponding blocks in different frames 940. This process helps augment the watermark signal present in the blocks, while reducing the noise common to the blocks. For example, merely adding blocks together may mitigate the effects of common noise.

[0162] In addition to common noise, other forms of noise may appear in each of the blocks such as noise introduced in the printing or scanning processes. Depending on the nature of the application, it may be advantageous to perform common noise recognition and removal at this stage 942. The developer may select a filter or series of filters to target certain types of noise that appear during experimentation with images. Certain types of median filters may be effective in mitigating the impact of spectral peaks (e.g., speckles) introduced in printing or scanning operations.

[0163] In addition to introducing noise, the printing and image capture processes may transform the color or orientation of the original, watermarked image. As described above, the embedder typically operates on a digital image in a particular color space and at a desired resolution. The watermark embedders normally operate on digital images represented in an RGB or CYMK color space at a desired resolution (e.g., 100 dpi or 300 dpi, the resolution at which the image is printed). The images are then printed on paper with a screen printing process that uses the CYMK subtractive color space at a line per inch (LPI) ranging from 65-200. 133 lines/in is typical for quality magazines and 73 lines/in is typical for newspapers. In order to produce a quality image and avoid pixelization, the rule of thumb is to use digital images with a resolution that is at least twice the press resolution. This is due to the half tone printing for color production. Also, different presses use screens with different patterns and line orientations and have different precision for color registration.

[0164] One way to counteract the transforms introduced through the printing process is to develop a model that characterizes these transforms and optimize watermark embedding and detecting based on this characterization. Such a model may be developed by passing watermarked and unwatermarked images through the printing process and observing the changes that occur to these images. The resulting model characterizes the changes introduced due to the printing process. The model may represent a transfer function that approximates the transforms due to the printing process. The detector then implements a pre-processing stage that reverses or at least mitigates the effect of the printing process on watermarked images. The detector may implement a pre-processing stage that performs the inverse of the transfer function for the printing process.

[0165] A related challenge is the variety in paper attributes used in different printing processes. Papers of various qualities, thickness and stiffness, absorb ink in various ways. Some papers absorb ink evenly, while others absorb ink at rates that vary with the changes in the paper's texture and thickness. These variations may degrade the embedded watermark signal when a digitally watermarked image is printed. The watermark process can counteract these effects by classifying and characterizing paper so that the embedder and reader can compensate for this printing-related degradation.

[0166] Variations in image capture processes also pose a challenge. In some applications, it is necessary to address problems introduced due to interlaced image data. Some video camera produce interlaced fields representing the odd or even scan lines of a frame. Problems arise when the interlaced image data consists of fields from two consecutive frames. To construct an entire frame, the preprocessor may combine the fields from consecutive frames while dealing with the distortion due to motion that occurs from one frame to the next. For example, it may be necessary to shift one field before interleaving it with another field to counteract interframe motion. A de-blurring function may be used to mitigate the blurring effect due to the motion between frames.

[0167] Another problem associated with cameras in some applications is blurring due to the lack of focus. The preprocessor can mitigate this effect by estimating parameters of a blurring function and applying a deblurring function to the input image.

[0168] Yet another problem associated with cameras is that they tend to have color sensors that utilize different color pattern implementations. As such, a sensor may produce colors slightly different than those represented in the object being captured. Most CCD and CMOS cameras use an array of sensors to produce colored images. The sensors in the array are arranged in clusters of sensitive to three primary colors red, green, and blue according to a specific pattern. Sensors designated for a particular color are dyed with that color to increase their sensitivity to the designated color. Many camera manufacturers use a Bayer color pattern GR/BG. While this pattern produces good image quality, it causes color mis-registration that degrades the watermark signal. Moreover, the color space converter, which maps the signal from the sensors to another color space such as YUV or RGB, may vary from one manufacturer to another. One way to counteract the mis-registration of the camera's color pattern is to account for the distortion due to the pattern in a color transformation process, implemented either within the camera itself, or as a pre-processing function in the detector.

[0169] Another challenge in counteracting the effects of the image capture process is dealing with the different types of distortion introduced from various image capture devices. For example, cameras have different sensitivities to light. In addition, their lenses have different spherical distortion, and noise characteristics. Some scanners have poor color reproduction or introduce distortion in the image aspect ratio. Some scanners introduce aliasing and employ interpolation to increase resolution. The detector can counteract these effects in the pre-processor by using an appropriate inverse transfer function. An off-line process first characterizes the distortion of several different image capture devices (e.g., by passing test images through the scanner and deriving a transfer function modeling the scanner distortion). Some detectors may be equipped with a library of such inverse transfer functions from which they select one that corresponds to the particular image capture device

[0170] Yet another challenge in applications where the image is printed on paper and later scanned is that the paper deteriorates over time and degrades the watermark. Also, varying lighting conditions make the watermark difficult to detect. Thus, the watermark may be selected so as to be more impervious to expected deterioration, and recoverable over a wider range of lighting conditions.

[0171] At the close of the pre-processing stage, the detector has selected a set of blocks for further processing. It then proceeds to gather evidence of the orientation signal in these blocks, and estimate the orientation parameters of promising orientation signal candidates. Since the image may have suffered various forms of corruption, the detector may identify several parts of the image that appear to have attributes similar to the orientation signal. As such, the detector may have to resolve potentially conflicting and ambiguous evidence of the orientation signal. To address this challenge, the detector estimates orientation parameters, and then refines theses estimates to extract the orientation parameters that are more likely to evince a valid signal than other parameter candidates.

4.2 Estimating Initial Orientation Parameters

[0172] FIG. 14 is a flow diagram illustrating a process for estimating rotation-scale vectors. The detector loops over each image block (950), calculating rotation-scale vectors with the best detection values in each block. First, the detector filters the block in a manner that tends to amplify the orientation signal while suppressing noise, including noise from the host image itself (952). Implemented as a multi-axis LaPlacian filter, the filter highlights edges (e.g., high frequency components of the image) and then suppresses them. The term, “multi-axis,” means that the filter includes a series of stages that each operates on particular axis. First, the filter operates on the rows of luminance samples, then operates on the columns, and adds the results. The filter may be applied along other axes as well. Each pass of the filter produces values at discrete levels. The final result is an array of samples, each having one of five values: {−2, −1, 0, 1, 2}.

[0173] Next, the detector performs a windowing operation on the block data to prepare it for an FFT transform (954). This windowing operation provides signal continuity at the block edges. The detector then performs an FFT (956) on the block, and retains only the magnitude component (958).

[0174] In an alternative implementation, the detector may use the phase signal produced by the FFT to estimate the translation parameter of the orientation signal. For example, the detector could use the rotation and scale parameters extracted in the process described below, and then compute the phase that provided the highest measure of correlation with the orientation signal using the phase component of the FFT process.

[0175] After computing the FFT, the detector applies a Fourier magnitude filter (960) on the magnitude components. The filter in the implementation slides over each sample in the Fourier magnitude array and filters the sample's eight neighbors in a square neighborhood centered at the sample. The filter boosts values representing a sharp peak with a rapid fall-off, and suppresses the fall-off portion. It also performs a threshold operation to clip peaks to an upper threshold.

[0176] Next, the detector performs a log-polar re-sample (962) of the filtered Fourier magnitude array to produce a log-polar array 964. This type of operation is sometimes referred to as a Fourier Mellin transform. The detector, or some off-line pre-processor, performs a similar operation on the orientation signal to map it to the log-polar coordinate system. Using matching filters, the detector implementation searches for a orientation signal in a specified window of the log-polar coordinate system. For example, consider that the log-polar coordinate system is a two dimensional space with the scale being the vertical axis and the angle being the horizontal axis. The window ranges from 0 to 90 degrees on the horizontal axis and from approximately 50 to 2400 dpi on the vertical axis. Note that the orientation pattern should be selected so that routine scaling does not push the orientation pattern out of this window. The orientation pattern can be designed to mitigate this problem, as noted above, and as explained in co-pending patent application No. 60/136,572, filed May 28, 1999, by Ammon Gustafson, entitled Watermarking System With Improved Technique for Detecting Scaling and Rotation, filed May 28, 1999.

[0177] The detector proceeds to correlate the orientation and the target signal in the log polar coordinate system. As shown in FIG. 14, the detector uses a generalized matched filter GMF (966). The GMF performs an FFT on the orientation and target signal, multiplies the resulting Fourier domain entities, and performs an inverse FFT. This process yields a rectangular array of values in log-polar coordinates, each representing a measure of correlation and having a corresponding rotation angle and scale vector. As an optimization, the detector may also perform the same correlation operations for distorted versions (968, 970, 972) of the orientation signal to see if any of the distorted orientation patterns results in a higher measure of correlation. For example, the detector may repeat the correlation operation with some pre-determined amount of horizontal and vertical differential distortion (970, 972). The result of this correlation process is an array of correlation values 974 specifying the amount of correlation that each corresponding rotation-scale vector provides.

[0178] The detector processes this array to find the top M peaks and their location in the log-polar space 976. To extract the location more accurately, the detector uses interpolation to provide the inter-sample location of each of the top peaks 978. The interpolator computes the 2D median of the samples around a peak and provides the location of the peak center to an accuracy of 0.1 sample.

[0179] The detector proceeds to rank the top rotation-scale vectors based on yet another correlation process 980. In particular, the detector performs a correlation between a Fourier magnitude representation for each rotation-scale vector candidate and a Fourier magnitude specification of the orientation signal 982. Each Fourier magnitude representation is scaled and rotated by an amount reflected by the corresponding rotation-scale vector. This correlation operation sums a point-wise multiplication of the orientation pattern impulse functions in the frequency domain with the Fourier magnitude values of the image at corresponding frequencies to compute a measure of correlation for each peak 984. The detector then sorts correlation values for the peaks (986).

[0180] Finally, the detector computes a detection value for each peak (988). It computes the detection value by quantizing the correlation values. Specifically, it computes a ratio of the peak's correlation value and the correlation value of the next largest peak. Alternatively, the detector may compute the ratio of the peak's correlation value and a sum or average of the correlation values of the next n highest peaks, where n is some predetermined number. Then, the detector maps this ratio to a detection value based on a statistical analysis of unmarked images.

[0181] The statistical analysis plots a distribution of peak ratio values found in unmarked images. The ratio values are mapped to a detection value based on the probability that the value came from an unmarked image. For example, 90% of the ratio values in unmarked images fall below a first threshold T1, and thus, the detection value mapping for a ratio of T1 is set to 1. Similarly, 99% of the ratio values in unmarked images fall below T2, and therefore, the detection value is set to 2. 99.9% of the ratio values in unmarked images fall below T3, and the corresponding detection value is set to 3. The threshold values, T1, T2 and T3, may be determined by performing a statistical analysis of several images. The mapping of ratios to detection values based on the statistical distribution may be implemented in a look up table.

[0182] The statistical analysis may also include a maximum likelihood analysis. In such an analysis, an off-line detector generates detection value statistics for both marked and unmarked images. Based on the probability distributions of marked and unmarked images, it determines the likelihood that a given detection value for an input image originates from a marked and unmarked image.

[0183] At the end of these correlation stages, the detector has computed a ranked set of rotation-scale vectors 990, each with a quantized measure of correlation associated with it. At this point, the detector could simply choose the rotation and scale vectors with the highest rank and proceed to compute other orientation parameters, such as differential scale and translation. Instead, the detector gathers more evidence to refine the rotation-scale vector estimates. FIG. 15 is a flow diagram illustrating a process for refining the orientation parameters using evidence of the orientation signal collected from blocks in the current frame.

[0184] Continuing in the current frame, the detector proceeds to compare the rotation and scale parameters from different blocks (e.g., block 0, block 1, block 2; 1000, 1002, and 1004 in FIG. 15). In a process referred to as interblock coincidence matching 1006, it looks for similarities between rotation-scale parameters that yielded the highest correlation in different blocks. To quantify this similarity, it computes the geometric distance between each peak in one block with every other peak in the other blocks. It then computes the probability that peaks will fall within this calculated distance. There are a variety of ways to calculate the probability. In one implementation, the detector computes the geometric distance between two peaks, computes the circular area encompassing the two peaks (π(geometric distance)²), and computes the ratio of this area to the total area of the block. Finally, it quantizes this probability measure for each pair of peaks (1008) by computing the log (base 10) of the ratio of the total area over the area encompassing the two peaks. At this point, the detector has calculated two detection values: quantized peak value, and the quantized distance metric.

[0185] The detector now forms multi-block grouping of rotation-scale vectors and computes a combined detection value for each grouping (1010). The detector groups vectors based on their relative geometric proximity within their respective blocks. It then computes the combined detection value by combining the detection values of the vectors in the group (1012). One way to compute a combined detection value is to add the detection values or add a weighted combination of them.

[0186] Having calculated the combined detection values, the detector sorts each grouping by its combined detection value (1014). This process produces a set of the top groupings of unrefined rotation-scale candidates, ranked by detection value 1016. Next, the detector weeds out rotation-scale vectors that are not promising by excluding those groupings whose combined detection values are below a threshold (the “refine threshold” 1018). The detector then refines each individual rotation-scale vector candidate within the remaining groupings.

[0187] The detector refines a rotation-scale vector by adjusting the vector and checking to see whether the adjustment results in a better correlation. As noted above, the detector may simply pick the best rotation-scale vector based on the evidence collected thus far, and refine only that vector. An alternative approach is to refine each of the top rotation-scale vector candidates, and continue to gather evidence for each candidate. In this approach, the detector loops over each vector candidate (1020), refining each one.

[0188] One approach of refining the orientation vector is as follows:

• • [0189] fix the orientation signal impulse functions (“points”) within a valid boundary (1022);
  • [0190] pre-refine the rotation-scale vector (1024);
  • [0191] find the major axis and re-fix the orientation points (1026); and
  • [0192] refine each vector with the addition of a differential scale component (1028).

[0193] In this approach, the detector pre-refines a rotation-scale vector by incrementally adjusting one of the parameters (scale, rotation angle), adjusting the orientation points, and then summing a point-wise multiplication of the orientation pattern and the image block in the Fourier magnitude domain. The refiner compares the resulting measure of correlation with previous measures and continues to adjust one of the parameters so long as the correlation increases. After refining the scale and rotation angle parameters, the refiner finds the major axis, and re-fixes the orientation points. It then repeats the refining process with the introduction of differential scale parameters. At the end of this process, the refiner has converted each scale-rotation candidate to a refined 4D vector, including rotation, scale, and two differential scale parameters.

[0194] At this stage, the detector can pick a 4D vector or set of 4D vector and proceed to calculate the final remaining parameter, translation. Alternatively, the detector can collect additional evidence about the merits of each 4D vector.

[0195] One way to collect additional evidence about each 4D vector is to re-compute the detection value of each orientation vector candidate (1030). For example, the detector may quantize the correlation value associated with each 4D vector as described above for the rotation-scale vector peaks (see item 988, FIG. 14 and accompanying text). Another way to collect additional evidence is to repeat the coincidence matching process for the 4D vectors. For this coincidence matching process, the detector computes spatial domain vectors for each candidate (1032), determines the distance metric between candidates from different blocks, and then groups candidates from different blocks based on the distance metrics (1034). The detector then re-sorts the groups according to their combined detection values (1036) to produce a set of the top P groupings 1038 for the frame.

[0196] FIG. 16 is a flow diagram illustrating a method for aggregating evidence of the orientation signal from multiple frames. In applications with multiple frames, the detector collects the same information for orientation vectors of the selected blocks in each frame (namely, the top P groupings of orientation vector candidates, e.g., 1050, 1052 and 1054). The detector then repeats coincidence matching between orientation vectors of different frames (1056). In particular, in this inter-frame mode, the detector quantizes the distance metrics computed between orientation vectors from blocks in different frames (1058). It then finds inter-frame groupings of orientation vectors (super-groups) using the same approach described above (1060), except that the orientation vectors are derived from blocks in different frames. After organizing orientation vectors into super-groups, the detector computes a combined detection value for each super-group (1062) and sorts the super-groups by this detection value (1064). The detector then evaluates whether to proceed to the next stage (1066), or repeat the above process of computing orientation vector candidates from another frame (1068).

[0197] If the detection values of one or more super-groups exceed a threshold, then the detector proceeds to the next stage. If not, the detector gathers evidence of the orientation signal from another frame and returns to the inter-frame coincidence matching process. Ultimately, when the detector finds sufficient evidence to proceed to the next stage, it selects the super-group with the highest combined detection value (1070), and sorts the blocks based on their corresponding detection values (1072) to produce a ranked set of blocks for the next stage (1074).

4.3 Estimating Translation Parameters

[0198] FIG. 17 is a flow diagram illustrating a method for estimating translation parameters of the orientation signal, using information gathered from the previous stages.

[0199] In this stage, the detector estimates translation parameters. These parameters indicate the starting point of a watermarked block in the spatial domain. The translation parameters, along with rotation, scale and differential scale, form a complete 6D orientation vector. The 6D vector enables the reader to extract luminance sample data in approximately the same orientation as in the original watermarked image.

[0200] One approach is to use generalized match filtering to find the translation parameters that provide the best correlation. Another approach is to continue to collect evidence about the orientation vector candidates, and provide a more comprehensive ranking of the orientation vectors based on all of the evidence gathered thus far. The following paragraphs describe an example of this type of an approach.

[0201] To extract translation parameters, the detector proceeds as follows. In the multi-frame case, the detector selects the frame that produced 4D orientation vectors with the highest detection values (1080). It then processes the blocks 1082 in that frame in the order of their detection value. For each block (1084), it applies the 4D vector to the luminance data to generate rectified block data (1086). The detector then performs dual axis filtering (1088) and the window function (1090) on the data. Next, it performs an FFT (1092) on the image data to generate an array of Fourier data. To make correlation operations more efficient, the detector buffers the fourier values at the orientation points (1094).

[0202] The detector applies a generalized match filter 1096 to correlate a phase specification of the orientation signal (1098) with the transformed block data. The result of this process is a 2D array of correlation values. The peaks in this array represent the translation parameters with the highest correlation. The detector selects the top peaks and then applies a median filter to determine the center of each of these peaks. The center of the peak has a corresponding correlation value and sub-pixel translation value. This process is one example of getting translation parameters by correlating the Fourier phase specification of the orientation signal and the image data. Other methods of phase locking the image data with a synchronization signal like the orientation signal may also be employed.

[0203] Depending on the implementation, the detector may have to resolve additional ambiguities, such as rotation angle and flip ambiguity. The degree of ambiguity in the rotation angle depends on the nature of the orientation signal. If the orientation signal is octally symmetric (symmetric about horizontal, vertical and diagonal axes in the spatial frequency domain), then the detector has to check each quadrant (0-90, 90-180, 180-270, and 270-360 degrees) to find out which one the rotation angle resides in. Similarly, if the orientation signal is quad symmetric, then the detector has to check two cases, 0-180 and 180-270.

[0204] The flip ambiguity may exist in some applications where the watermarked image can be flipped. To check for rotation and flip ambiguities, the detector loops through each possible case, and performs the correlation operation for each one (1100).

[0205] At the conclusion of the correlation process, the detector has produced a set of the top translation parameters with associated correlation values for each block. To gather additional evidence, the detector groups similar translation parameters from different blocks (1102), calculates a group detection value for each set of translation parameters 1104, and then ranks the top translation groups based on their corresponding group detection values 1106.

4.4 Refining Translation Parameters

[0206] Having gathered translation parameter estimates, the detector proceeds to refine these estimates. FIG. 18 is a flow diagram illustrating a process for refining orientation parameters. At this stage, the detector process has gathered a set of the top translation parameter candidates 1120 for a given frame 1122. The translation parameters provide an estimate of a reference point that locates the watermark, including both the orientation and message components, in the image frame. In the implementation depicted here, the translation parameters are represented as horizontal and vertical offsets from a reference point in the image block from which they were computed.

[0207] Recall that the detector has grouped translation parameters from different blocks based on their geometric proximity to each other. Each pair of translation parameters in a group is associated with a block and a 4D vector (rotation, scale, and 2 differential scale parameters). As shown in FIG. 18, the detector can now proceed to loop through each group (1124), and through the blocks within each group (1126), to refine the orientation parameters associated with each member of the groups. Alternatively, a simpler version of the detector may evaluate only the group with the highest detection value, or only selected blocks within that group.

[0208] Regardless of the number of candidates to be evaluated, the process of refining a given orientation vector candidate may be implemented in a similar fashion. In the refining process, the detector uses a candidate orientation vector to define a mesh of sample blocks for further analysis (1128). In one implementation, for example, the detector forms a mesh of 32 by 32 sample blocks centered around a seed block whose upper right corner is located at the vertical and horizontal offset specified by the candidate translation parameters. The detector reads samples from each block using the orientation vector to extract luminance samples that approximate the original orientation of the host image at encoding time.

[0209] The detector steps through each block of samples (1130). For each block, it sets the orientation vector (1132), and then uses the orientation vector to check the validity of the watermark signal in the sample block. It assesses the validity of the watermark signal by calculating a figure of merit for the block (1134). To further refine the orientation parameters associated with each sample block, the detector adjusts selected parameters (e.g., vertical and horizontal translation) and re-calculates the figure of merit. As depicted in the inner loop in FIG. 18 (block 1136 to 1132), the detector repeatedly adjusts the orientation vector and calculates the figure of merit in an attempt to find a refined orientation that yields a higher figure of merit.

[0210] The loop (1136) may be implemented by stepping through a predetermined sequence of adjustments to parameters of the orientation vectors (e.g., adding or subtracting small increments from the horizontal and vertical translation parameters). In this approach, the detector exits the loop after stepping through the sequence of adjustments. Upon exiting, the detector retains the orientation vector with the highest figure of merit.

[0211] There are a number of ways to calculate this figure of merit. One figure of merit is the degree of correlation between a known watermark signal attribute and a corresponding attribute in the signal suspected of having a watermark. Another figure of merit is the strength of the watermark signal (or one of its components) in the suspect signal. For example, a figure of merit may be based on a measure of the watermark message signal strength and/or orientation pattern signal strength in the signal, or in a part of the signal from which the detector extracts the orientation parameters. The detector may computes a figure of merit based the strength of the watermark signal in a sample block. It may also compute a figure of merit based on the percentage agreement between the known bits of the message and the message bits extracted from the sample block.

[0212] When the figure of merit is computed based on a portion of the suspect signal, the detector and reader can use the figure of merit to assess the accuracy of the watermark signal detected and read from that portion of the signal. This approach enables the detector to assess the merits of orientation parameters and to rank them based on their figure of merit. In addition, the reader can weight estimates of watermark message values based on the figure of merit to recover a message more reliably.

[0213] The process of calculating a figure of merit depends on attributes the watermark signal and how the embedder inserted it into the host signal. Consider an example where the watermark signal is added to the host signal. To calculate a figure of merit based on the strength of the orientation signal, the detector checks the value of each sample relative to its neighbors, and compares the result with the corresponding sample in a spatial domain version of the orientation signal. When a sample's value is greater than its neighbors, then one would expect that the corresponding orientation signal sample to be positive. Conversely, when the sample's value is less than its neighbors, then one would expect that the corresponding orientation sample to be negative. By comparing a sample's polarity relative to its neighbors with the corresponding orientation sample's polarity, the detector can assess the strength of the orientation signal in the sample block. In one implementation, the detector makes this polarity comparison twice for each sample in an N by N block (e.g., N=32, 64, etc): once comparing each sample with its horizontally adjacent neighbors and then again comparing each sample with its vertically adjacent neighbors. The detector performs this analysis on samples in the mesh block after reorienting the data to approximate the original orientation of the host image at encoding time. The result of this process is a number reflecting the portion of the total polarity comparisons that yield a match.

[0214] To calculate a figure of merit based on known signature bits in a message, the detector invokes the reader on the sample block, and provides the orientation vector to enable the reader to extract coded message bits from the sample block. The detector compares the extracted message bits with the known bits to determine the extent to which they match. The result of this process is a percentage agreement number reflecting the portion of the extracted message bits that match the known bits. Together the test for the orientation signal and the message signal provide a figure of merit for the block.

[0215] As depicted in the loop from blocks 1138 to 1130, the detector may repeat the process of refining the orientation vector for each sample block around the seed block. In this case, the detector exits the loop (1138) after analyzing each of the sample blocks in the mesh defined previously (1128). In addition, the detector may repeat the analysis in the loop through all blocks in a given group (1140), and in the loop through each group (1142).

[0216] After completing the analysis of the orientation vector candidates, the detector proceeds to compute a combined detection value for the various candidates by compiling the results of the figure of merit calculations. It then proceeds to invoke the reader on the orientation vector candidates in the order of their detection values.

4.5 Reading the Watermark

[0217] FIG. 19 is a flow diagram illustrating a process for reading the watermark message. Given an orientation vector and the corresponding image data, the reader extracts the raw bits of a message from the image. The reader may accumulate evidence of the raw bit values from several different blocks. For example, in the process depicted in FIG. 19, the reader uses refined orientation vectors for each block, and accumulates evidence of the raw bit values extracted from the blocks associated with the refined orientation vectors.

[0218] The reading process begins with a set of promising orientation vector candidates 1150 gathered from the detector. In each group of orientation vector candidates, there is a set of orientation vectors, each corresponding to a block in a given frame. The detector invokes the reader for one or more orientation vector groups whose detection values exceed a predetermined threshold. For each such group, the detector loops over the blocks in the group (1152), and invokes the reader to extract evidence of the raw message bit values.

[0219] Recall that previous stages in the detector have refined orientation vectors to be used for the blocks of a group. When it invokes the reader, the detector provides the orientation vector as well as the image block data (1154). The reader scans samples starting from a location in a block specified by the translation parameters and using the other orientation parameters to approximate the original orientation of the image data (1156).

[0220] As described above, the embedder maps chips of the raw message bits to each of the luminance samples in the original host image. Each sample, therefore, may provide an estimate of a chip's value. The reader reconstructs the value of the chip by first predicting the watermark signal in the sample from the value of the sample relative to its neighbors as described above (1158). If the deduced value appears valid, then the reader extracts the chip's value using the known value of the pseudo-random carrier signal for that sample and performing the inverse of the modulation function originally used to compute the watermark information signal (1160). In particular, the reader performs an exclusive OR operation on the deduced value and the known carrier signal bit to get an estimate of the raw bit value. This estimate serves as an estimate for the raw bit value. The reader accumulates these estimates for each raw bit value (1162).

[0221] As noted above, the reader computes an estimate of the watermark signal by predicting the original, un-watermarked signal and deriving an estimate of the watermark signal based on the predicted signal and the watermarked signal. It then computes an estimate of a raw bit value based on the value of the carrier signal, the assignment map that maps a raw bit to the host image, and the relationship among the carrier signal value, the raw bit value, and the watermark signal value. In short, the reader reverses the embedding functions that modulate the message with the carrier and apply the modulated carrier to the host signal. Using the predicted value of the original signal and an estimate of the watermark signal, the reader reverses the embedding functions to estimate a value of the raw bit.

[0222] The reader loops over the candidate orientation vectors and associated blocks, accumulating estimates for each raw bit value (1164). When the loop is complete, the reader calculates a final estimate value for each raw bit from the estimates compiled for it. It then performs the inverse of the error correction coding operation on the final raw bit values (1166). Next, it performs a CRC to determine whether the read is valid. If no errors are detected, the read operation is complete and the reader returns the message (1168).

[0223] However, if the read is invalid, then the detector may either attempt to refine the orientation vector data further, or start the detection process with a new frame. Preferably, the detector should proceed to refine the orientation vector data when the combined detection value of the top candidates indicates that the current data is likely to contain a strong watermark signal. In the process depicted in FIG. 19, for example, the detector selects a processing path based on the combined detection value (1170). The combined detection value may be calculated in a variety of ways. One approach is to compute a combined detection value based on the geometric coincidence of the top orientation vector candidates and a compilation of their figures of merit. The figure of merit may be computed as detailed earlier.

[0224] For cases where the read is invalid, the processing paths for the process depicted in FIG. 19 include: 1) refine the top orientation vectors in the spatial domain (1172); 2) invoke the translation estimator on the frame with the next best orientation vector candidates (1174); and 3) re-start the detection process on a new frame (assuming an implementation where more than one frame is available) (1176). These paths are ranked in order from the highest detection value to the lowest. In the first case, the orientation vectors are the most promising. Thus, the detector re-invokes the reader on the same candidates after refining them in the spatial domain (1178). In the second case, the orientation vectors are less promising, yet the detection value indicates that it is still worthwhile to return to the translation estimation stage and continue from that point. Finally, in the final case, the detection value indicates that the watermark signal is not strong enough to warrant further refinement. In this case, the detector starts over with the next new frame of image data.

[0225] In each of the above cases, the detector continues to process the image data until it either makes a valid read, or has failed to make a valid read after repeated passes through the available image data.

5.0 Operating Environment for Computer Implementations

[0226] FIG. 20 illustrates an example of a computer system that serves as an operating environment for software implementations of the watermarking systems described above. The embedder and detector implementations are implemented in C/C++ and are portable to many different computer systems. FIG. 20 generally depicts one such system.

[0227] The computer system shown in FIG. 20 includes a computer 1220, including a processing unit 1221, a system memory 1222, and a system bus 1223 that interconnects various system components including the system memory to the processing unit 1221.

[0228] The system bus may comprise any of several types of bus structures including a memory bus or memory controller, a peripheral bus, and a local bus using a bus architecture such as PCI, VESA, Microchannel (MCA), ISA and EISA, to name a few.

[0229] The system memory includes read only memory (ROM) 1224 and random access memory (RAM) 1225. A basic input/output system 1226 (BIOS), containing the basic routines that help to transfer information between elements within the computer 1220, such as during start-up, is stored in ROM 1224.

[0230] The computer 1220 further includes a hard disk drive 1227, a magnetic disk drive 1228, e.g., to read from or write to a removable disk 1229, and an optical disk drive 1230, e.g., for reading a CD-ROM or DVD disk 1231 or to read from or write to other optical media. The hard disk drive 1227, magnetic disk drive 1228, and optical disk drive 1230 are connected to the system bus 1223 by a hard disk drive interface 1232, a magnetic disk drive interface 1233, and an optical drive interface 1234, respectively. The drives and their associated computer-readable media provide nonvolatile storage of data, data structures, computer-executable instructions (program code such as dynamic link libraries, and executable files), etc. for the computer 1220.

[0231] Although the description of computer-readable media above refers to a hard disk, a removable magnetic disk and an optical disk, it can also include other types of media that are readable by a computer, such as magnetic cassettes, flash memory cards, digital video disks, and the like.

[0232] A number of program modules may be stored in the drives and RAM 1225, including an operating system 1235, one or more application programs 1236, other program modules 1237, and program data 1238.

[0233] A user may enter commands and information into the computer 1220 through a keyboard 1240 and pointing device, such as a mouse 1242. Other input devices may include a microphone, joystick, game pad, satellite dish, digital camera, scanner, or the like. A digital camera or scanner 43 may be used to capture the target image for the detection process described above. The camera and scanner are each connected to the computer via a standard interface 44. Currently, there are digital cameras designed to interface with a Universal Serial Bus (USB), Peripheral Component Interconnect (PCI), and parallel port interface. Two emerging standard peripheral interfaces for cameras include USB2 and 1394 (also known as firewire and iLink).

[0234] Other input devices may be connected to the processing unit 1221 through a serial port interface 1246 or other port interfaces (e.g., a parallel port, game port or a universal serial bus (USB)) that are coupled to the system bus.

[0235] A monitor 1247 or other type of display device is also connected to the system bus 1223 via an interface, such as a video adapter 1248. In addition to the monitor, computers typically include other peripheral output devices (not shown), such as speakers and printers.

[0236] The computer 1220 operates in a networked environment using logical connections to one or more remote computers, such as a remote computer 1249. The remote computer 1249 may be a server, a router, a peer device or other common network node, and typically includes many or all of the elements described relative to the computer 1220, although only a memory storage device 1250 has been illustrated in FIG. 20. The logical connections depicted in FIG. 20 include a local area network (LAN) 1251 and a wide area network (WAN) 1252. Such networking environments are commonplace in offices, enterprise-wide computer networks, intranets and the Internet.

[0237] When used in a LAN networking environment, the computer 1220 is connected to the local network 1251 through a network interface or adapter 1253. When used in a WAN networking environment, the computer 1220 typically includes a modem 1254 or other means for establishing communications over the wide area network 1252, such as the Internet. The modem 1254, which may be internal or external, is connected to the system bus 1223 via the serial port interface 1246.

[0238] In a networked environment, program modules depicted relative to the computer 1220, or portions of them, may be stored in the remote memory storage device. The processes detailed above can be implemented in a distributed fashion, and as parallel processes. It will be appreciated that the network connections shown are exemplary and that other means of establishing a communications link between the computers may be used.

[0239] While the computer architecture depicted in FIG. 20 is similar to typical personal computer architectures, aspects of the invention may be implemented in other computer architectures, such as hand-held computing devices like Personal Digital Assistants, audio and/video players, network appliances, telephones, etc.

[0240] Watermark Processing of Images, Printed Images, and Security Documents

[0241] Digital watermarking (sometimes termed “data hiding” or “data embedding”) is a growing field of endeavor, with several different approaches. The present assignee's work is reflected in the patents and applications detailed above, together with laid-open PCT application WO97/43736. Other work is illustrated by U.S. Pat. Nos. 5,734,752, 5,646,997, 5,659,726, 5,664,018, 5,671,277, 5,687,191, 5,687,236, 5,689,587, 5,568,570, 5,572,247, 5,574,962, 5,579,124, 5,581,500, 5,613,004, 5,629,770, 5,461,426, 5,743,631, 5,488,664, 5,530,759, 5,539,735, 4,943,973, 5,337,361, 5,404,160, 5,404,377, 5,315,098, 5,319,735, 5,337,362, 4,972,471, 5,161,210, 5,243,423, 5,091,966, 5,113,437, 4,939,515, 5,374,976, 4,855,827, 4,876,617, 4,939,515, 4,963,998, 4,969,041, and published foreign applications WO 98/02864, EP 822,550, WO 97/39410, WO 96/36163, GB 2,196,167, EP 777,197, EP 736,860, EP 705,025, EP 766,468, EP 782,322, WO 95/20291, WO 96/26494, WO 96/36935, WO 96/42151, WO 97/22206, WO 97/26733. Some of the foregoing patents relate to visible watermarking techniques. Other visible watermarking techniques (e.g. data glyphs) are described in U.S. Pat. Nos. 5,706,364, 5,689,620, 5,684,885, 5,680,223, 5,668,636, 5,640,647, 5,594,809.

[0242] Much of the work in data embedding is not in the patent literature but rather is published in technical articles. In addition to the patentees of the foregoing patents, some of the other workers in this field (whose watermark-related writings can by found by an author search in the INSPEC or NEXIS databases, among others) include I. Pitas, Eckhard Koch, Jian Zhao, Norishige Morimoto, Laurence Boney, Kineo Matsui, A. Z. Tirkel, Fred Mintzer, B. Macq, Ahmed H. Tewfik, Frederic Jordan, Naohisa Komatsu, Joseph O'Ruanaidh, Neil Johnson, Ingemar Cox, Minerva Yeung, and Lawrence O'Gorman.

[0243] The artisan is assumed to be familiar with the foregoing prior art.

[0244] In the following disclosure it should be understood that references to watermarking encompass not only the assignee's watermarking technology, but can likewise be practiced with any other watermarking technology, such as those indicated above.

[0245] Watermarking can be applied to myriad forms of information. The present disclosure focuses on its applications to security documents. However, it should be recognized that the principles discussed below could also be applied outside this area.

[0246] Most of the prior art in image watermarking has focused on pixelated imagery (e.g. bit-mapped images, JPEG/MPEG imagery, VGA/SVGA display devices, etc.). In most watermarking techniques, the luminance or color values of component pixels are slightly changed to effect subliminal encoding of binary data through the image. (This encoding can be done directly in the pixel domain, or after the signal has been processed and represented differently—e.g. as DCT or wavelet coefficients, or as compressed data, etc.)

[0247] While pixelated imagery is a relatively recent development, security documents—commonly employing line art—go back centuries. One familiar example is U.S. paper currency. On the one dollar banknote, for example, line art is used in several different ways. One is to form intricate webbing patterns (sometimes termed “guilloche patterns”) around the margin of the note (generally comprised of light lines on dark background). Another is to form gray scale imagery, such as the portrait of George Washington (generally comprised of dark lines on a light background).

[0248] There are two basic ways to simulate grey-scales in security document line art. One is to change the relative spacings of the lines to effect a lightening or darkening of an image region. FIG. 21A shows such an arrangement; area B looks darker than area A due to the closer spacings of the component lines. The other technique is to change the widths of the component lines—wider lines resulting in darker areas and narrower lines resulting in lighter areas. FIG. 21B shows such an arrangement. Again, area B looks darker than area A, this time due to the greater widths of the component lines. These techniques are often used together. Ultimately, a given region simply has more or less ink.

[0249] In my application Ser. No. 08/438,159 (now U.S. Pat. No. 5,850,481), I introduced, and in my application Ser. No. 09/074,034 (now U.S. Pat. No. 6,449,377) I elaborated on, techniques for watermarking line art by making slight changes to the widths, or positions, of the component lines. Such techniques are further expanded in the present disclosure.

[0250] In several of my cited applications, I discussed various “calibration signals” that can be used to facilitate the decoding of watermark data despite corruption of the encoded image, such as by scaling or rotation. Common counterfeiting techniques—e.g. color photocopying, or scanning/inkjet printing—often introduce such corruption, whether deliberately or accidentally. Accordingly, it is important that watermarks embedded in security documents be detectable notwithstanding such effects. Calibration signals particularly suited for use with security documents are detailed in this disclosure.

[0251] In some embodiments, security documents are encoded to convey machine-readable multi-bit binary information (e.g. digital watermarks), usually in a manner not alerting human viewers that such information is present. The documents can be provided with overt or subliminal calibration patterns. When a document incorporating such a pattern is scanned (e.g. by a photocopier), the pattern facilitates detection of the encoded information notwithstanding possible scaling or rotation of the scan data. The calibration pattern can serve as a carrier for the watermark information, or the watermark can be encoded independently. In one embodiment, the watermark and the calibration pattern are formed on the document by an intaglio process, with or without ink. A photocopier responsive to such markings can take predetermined action if reproduction of a security document is attempted. A passport processing station responsive to such markings can use the decoded binary data to access a database having information concerning the passport holder. Some such apparatuses detect both the watermark data and the presence of a visible structure characteristic of a security document (e.g., the seal of the issuing central bank).

[0252] By way of introduction, the present specification begins with review of techniques for embedding watermark data in line art, as disclosed in my application Ser. No. 09/074,034 (now U.S. Pat. No. 6,449,377).

[0253] Referring to FIG. 22, the earlier-described technique employs a grid 10 of imaginary reference points arrayed over a line art image. The spacing between points is 250 microns in the illustrated arrangement, but greater or lesser spacings can of course be used.

[0254] Associated with each grid point is a surrounding region 12, shown in FIG. 23. As described below, the luminosity (or reflectance) of each of these regions 12 is slightly changed to effect subliminal encoding of binary data.

[0255] Region 12 can take various shapes; the illustrated rounded-rectangular shape is representative only. (The illustrated shape has the advantage of encompassing a fairly large area while introducing fewer visual artifacts than, e.g., square regions.) In other embodiments, squares, rectangles, circles, ellipses, etc., can alternatively be employed.

[0256] FIG. 24 is a magnified view of an excerpt of FIG. 23, showing a line 14 passing through the grid of points. The width of the line, of course, depends on the particular image of which it is a part. The illustrated line is about 40 microns in width; greater or lesser widths can naturally be used.

[0257] In one encoding technique, shown in FIG. 25, the width of the line is controllably varied so as to change the luminosity of the regions through which it passes. To increase the luminosity (or reflectance), the line is made narrower (i.e. less ink in the region). To decrease the luminosity, the line is made wider (i.e. more ink).

[0258] Whether the luminance in a given region should be increased or decreased depends on the particular watermarking algorithm used. Any algorithm can be used, by changing the luminosity of regions 12 as the algorithm would otherwise change the luminance or colors of pixels in a pixelated image. (Some watermarking algorithms effect their changes in a transformed domain, such as DCT, wavelet, or Fourier. However, such changes are ultimately manifested as changes in luminance or color.)

[0259] In an exemplary algorithm, the binary data is represented as a sequence of −1s and 1s, instead of 0s and 1s. (The binary data can comprise a single datum, but more typically comprises several. In an illustrative embodiment, the data comprises 128 bits, some of which are error-correcting or -detecting bits.)

[0260] Each element of the binary data sequence is then multiplied by a corresponding element of a pseudorandom number sequence, comprised of −1s and 1s, to yield an intermediate data signal. Each element of this intermediate data signal is mapped to a corresponding sub-part of the image, such as a region 12. (Commonly, each element is mapped to several such sub-parts.) The image in (and optionally around) this region is analyzed to determine its relative capability to conceal embedded data, and a corresponding scale factor is produced. Exemplary scale factors may range from 0 to 3. The scale factor for the region is then multiplied by the element of the intermediate data signal mapped to the region in order to yield a “tweak” or “bias” value for the region. In the illustrated case, the resulting tweaks can range from −3 to 3. The luminosity of the region is then adjusted in accordance with the tweak value. A tweak value of −3 may correspond to a −5% change in luminosity; −2 may correspond to −2% change; −1 may correspond to −1% change; 0 may correspond to no change; 1 may correspond to +1% change; 2 may correspond to +2% change, and 3 may correspond to +5% change. (This example follows the basic techniques described in the Real Time Encoder embodiment disclosed in U.S. Pat. No. 5,710,834, which is hereby incorporated by reference.)

[0261] In FIG. 25, the watermarking algorithm determined that the luminance of region A should be reduced by a certain percentage, while the luminance of regions C and D should be increased by certain percentages.

[0262] In region A, the luminance is reduced by increasing the line width. In region D, the luminance is increased by reducing the line width; similarly in region C (but to a lesser extent).

[0263] No line passes through region B, so there is no opportunity to change the region's luminance. This is not fatal to the method, however, since the exemplary watermarking algorithm redundantly encodes each bit of data in sub-parts spaced throughout the line art image.

[0264] The changes to line widths in regions A and D of FIG. 25 are exaggerated for purposes of illustration. While the illustrated variance is possible, most implementations will typically modulate the line width 3-50% (increase or decrease).

[0265] (Many watermarking algorithms routinely operate within a signal margin of about +/−1% changes in luminosity to effect encoding. That is, the “noise” added by the encoding amounts to just 1% or so of the underlying signal. Lines typically don't occupy the full area of a region, so a 10% change to line width may only effect a 1% change to region luminosity, etc. Security documents are different from photographs in that the artwork generally need not convey photorealism. Thus, security documents can be encoded with higher energy than is used in watermarking photographs, provided the result is still aesthetically satisfactory. To illustrate, localized luminance changes on the order of 10% are possible in security documents, while such a level of watermark energy in photographs would generally be considered unacceptable. In some contexts, localized luminance changes of 20, 30, 50 or even 100% are acceptable.)

[0266] In the illustrated technique, the change to line width is a function solely of the watermark tweak (or watermark/calibration pattern tweak, as discussed below) to be applied to a single region. Thus, if a line passes through any part of a region to which a tweak of 2% is to be applied, the line width in that region is changed to effect the 2% luminance difference. In variant techniques, the change in line width is a function of the line's position in the region. In particular, the change in line width is a function of the distance between the region's center grid point and the line's closest approach to that point. If the line passes through the grid point, the full 2% change is effected. At successively greater distances, successively smaller changes are applied. The manner in which the magnitude of the tweak changes as a function of line position within the region can be determined by applying one of various interpolation algorithms, such as the bi-linear, bi-cubic, cubic splines, custom curve, etc.

[0267] In other variant techniques, the change in line width in a given region is a weighted function of the tweaks for adjoining or surrounding regions. Thus, the line width in one region may be increased or decreased in accordance with a tweak value corresponding to one or more adjoining regions.

[0268] Combinations of the foregoing techniques can also be employed.

[0269] In the foregoing techniques, it is sometimes necessary to trade-off the tweak values of adjoining regions. For example, a line may pass along a border between regions, or pass through the point equidistant from four grid points (“equidistant zones”). In such cases, the line may be subject to conflicting tweak values—one region may want to increase the line width, while another may want to decrease the line width. (Or both may want to increase the line width, but differing amounts.) Similarly in cases where the line does not pass through an equidistant zone, but the change in line width is a function of a neighborhood of regions whose tweaks are of different values. Again, known interpolation functions can be employed to determine the weight to be given the tweak from each region in determining what change is to be made to the line width in any given region.

[0270] In the exemplary watermarking algorithm, the average change in luminosity across the security document image is zero, so no generalized lightening or darkening of the image is apparent. The localized changes in luminosity are so minute in magnitude, and localized in position, that they are essentially invisible (e.g. inconspicuous/subliminal) to human viewers.

[0271] An alternative technique is shown in FIG. 26, in which line position is changed rather than line width.

[0272] In FIG. 26 the original position of the line is shown in dashed form, and the changed position of the line is shown in solid form. To decrease a region's luminosity, the line is moved slightly closer to the center of the grid point; to increase a region's luminosity, the line is moved slightly away. Thus, in region A, the line is moved towards the center grid point, while in region D it is moved away.

[0273] It will be noted that the line on the left edge of region A does not return to its nominal (dashed) position as it exits the region. This is because the region to the left of region A also is to have decreased luminosity. Where possible, it is generally preferable not to return a line to its nominal position, but instead to permit shifted lines to remain shifted as they enter adjoining regions. So doing permits a greater net line movement within a region, increasing the embedded signal level.

[0274] Again, the line shifts in FIG. 26 are somewhat exaggerated. More typical line shifts are on the order of 3-50 microns.

[0275] One way to think of the FIG. 26 technique is to employ a magnetism analogy. The grid point in the center of each region can be thought of as a magnet. It either attracts or repels lines. A tweak value of −3, for example, may correspond to a strong-valued attraction force; a tweak value of +2 may correspond to a middle-valued repulsion force, etc. In FIG. 26, the grid point in region A exhibits an attraction force (i.e. a negative tweak value), and the grid point in region D exhibits a repulsion force (e.g. a positive tweak value).

[0276] The magnetic analogy is useful because the magnetic effect exerted on a line depends on the distance between the line and the grid point. Thus, a line passing near a grid point is shifted more in position than a line near the periphery of the region.

[0277] (Actually, the magnetism analogy can serve as more than a conceptual tool. Instead, magnetic effects can be modeled in a computer program and serve to synthesize a desired placement of the lines relative to the grid points. Arbitrarily customized magnetic fields can be used.)

[0278] Each of the variants applicable to FIG. 25 is likewise applicable to FIG. 26.

[0279] Combinations of the embodiments of FIGS. 25 and 26 can of course be used, resulting in increased watermark energy, better signal-to-noise ratio and, in many cases, less noticeable changes.

[0280] In still a further technique, the luminance in each region is changed while leaving the line unchanged. This can be effected by sprinkling tiny dots of ink in the otherwise-vacant parts of the region. In high quality printing, of the type used with security documents, droplets on the order of 3 microns in diameter can be deposited. (Still larger droplets are still beyond the perception threshold for most viewers.) Speckling a region with such droplets (either in a regular array, or random, or according to a desired profile such as Gaussian), can readily effect a 1% or so change in luminosity. (Usually dark droplets are added to a region, effecting a decrease in luminosity. Increases in luminosity can be effected by speckling with a light colored ink, or by forming light voids in line art otherwise present in a region.) (Actually, production realities often mean that many such microdots will not print, but statistically some will.)

[0281] In a variant of the speckling technique, very thin mesh lines can be inserted in the artwork—again to slightly change the luminance of one or more regions (so-called “background tinting”).

[0282] The following portion of the specification reviews a calibration, or synchronization pattern used in an illustrative security document to facilitate proper registration of the watermark data for decoding. It may be helpful to being by reviewing further details about the illustrative watermarking method.

[0283] Referring to FIG. 27A, an exemplary watermark is divided into “cells” that are 250 microns on a side, each conveying a single bit of information. The cells are grouped into a “block” having 128 cells on a side (i.e. 16,384 cells per block). The blocks are tiled across the region being watermarked (e.g. across the face of a security document).

[0284] As noted, the watermark payload consists of 128 bits of data. Each bit is represented by 128 different cells within each block. (The mapping of bits to cells can be pseudo-random, sequential, or otherwise.) The 128 “0”s and “1”s of the watermark data are randomized into substantially equal-probability “1”s and “−1”s by a pseudo-random function to reduce watermark visibility. Where a cell has a value of “1,” the luminance of the corresponding area of the image is slightly increased; where a cell has a value of “−1,” the luminance of the corresponding area of the image is slightly decreased (or vice versa). In some embodiments, the localized changes to image luminance due to the +1/−1 watermark cell values are scaled in accordance with data-hiding attributes of the local area (e.g. to a range of +/−4 digital numbers) to increase the robustness of the watermark without compromising its imperceptibility.

[0285] It should be noted that a single watermark “cell” commonly encompasses a large number of ink droplets. In high resolution printing, as is commonly used in security documents (e.g. 5000 microdroplets per inch), a single watermark cell may encompass a region of 50 droplets by 50 droplets. In other embodiments, a cell may encompass greater or lesser numbers of droplets.

[0286] Decoding a watermark requires precise re-registration of the scanned document image, so the watermark cells are located where expected. To facilitate such registration, a calibration signal can be employed.

[0287] An exemplary calibration signal is a geometrical pattern having a known Fourier-Mellin transform. As described in application Ser. No. 08/649,419 (now U.S. Pat. No. 5,862,260), when a known pattern is transformed into the Fourier domain, and then further transformed into the Fourier-Mellin domain, the transformed data indicates the scale and rotation of the pattern. If this pattern is replicated on a security document that is thereafter scanned (as noted, scanning commonly introduces rotation, and sometimes scaling), the F-M transform data indicates the scale and rotation of the scanned data, facilitating virtual re-registration of the security document image for watermark detection.

[0288] As shown in FIG. 27B, an illustrative geometrical calibration pattern is a block, 3.2 cm on a side. The block comprises a 16×16 array of substantially identical tiles, each 2 mm on a side. Each tile, in term, comprises an 8×8 array of component cells.

[0289] As described below, the geometrical calibration pattern in the illustrated embodiment is a visible design feature on the security document. Accordingly, unlike the watermark data, the calibration pattern does not have to be limited to a small range of digital numbers in order to keep it substantially hidden among other features of the document. Also unlike the watermark data, the illustrated calibration pattern is not locally scaled in accordance with data hiding attributes of the security document image.

[0290] It is possible to print rectangular grids of grey-scaled ink on a document to serve as a calibration pattern. However, aesthetic considerations usually discourage doing so. Preferable is to realize the calibration pattern in a more traditional art form, such as a seemingly random series of intertwining lines, forming a weave-like pattern that is printed across part or all of the document.

[0291] To create this weave-like calibration pattern, a designer first defines an 8×8 cell reference calibration tile. Each cell in the tile is assigned a grey-scale value. In the illustrated embodiment, values within 2-10 percent of each other are used, although this is not essential. An exemplary reference calibration tile is shown in FIG. 28 (assuming 8-bit quantization).

[0292] The Fourier-Mellin transform of a block derived from this reference calibration tile will serve as the key by which the scale and rotation of a scanned security document image are determined.

[0293] There is some optimization that may be done in selecting/designing the pattern of grey-scale values that define the reference calibration tile. The pattern should have a F-M transform that is readily distinguished from those of other design and watermark elements on the security document. One design procedure effects a trial F-M transform of the rest of the security document design, and works backwards from this data to select a reference calibration tile that is readily distinguishable.

[0294] Once a reference tile pattern is selected, the next steps iteratively define a tile having a weave-like pattern whose local luminance values approximately match the reference tile's grey-scale pattern.

[0295] Referring to FIG. 29A, the first such step is to select points on the bottom and left side edges of the tile where lines are to cross the tile boundaries. The angles at which the lines cross these boundaries are also selected. (In the illustrated embodiment, these points and angles are selected arbitrarily, although in other embodiments, the choices can be made in conformance with an optimizing design procedure.)

[0296] The selected points and angles are then replicated on the corresponding right and top edges of the tile. By this arrangement, lines exiting the top of one tile seamlessly enter the bottom of the adjoining tile at the same angle. Likewise, lines exiting either side of a tile seamlessly join with lines in the laterally adjoining blocks.

[0297] The designer next establishes trial line paths snaking through the tile (FIGS. 29B, 29C), linking arbitrarily matched pairs of points on the tile's edges. (These snaking paths are sometimes termed “worms.”) Desirably, these paths pass through each of the 64 component cells forming the tile, with the total path length through each cell being within +/−30% of the average path length through all cells. (This trial routing can be performed with pencil and paper, but more commonly is done on a computer graphics station, with a mouse, light pen, or other input device being manipulated by the designer to establish the routing.) In the illustrated embodiment, the lines have a width of about 30-100 microns, and an average spacing between lines of about 100-400 microns, although these parameters are not critical.

[0298] Turning next to FIG. 30, the trial tile is assembled with like tiles to form a 16×16 trial block (3.2 cm on a side), with a repetitive weave pattern formed by replication of the line pattern defined on the 8×8 cell trial tile. This trial block is then converted into grey-scale values. The conversion can be done by scanning a printed representation of the trial block, or by computer analysis of the line lengths and positions. The output is a 128×128 array of grey-scale values, each value corresponding to the luminance of a 250 micron cell within the trial block.

[0299] This grey-scale data is compared with grey-scale data provided by assembling 256 of the reference calibration tiles (each an 8×8 array of cells) into a 16×16 calibration pattern block. In particular, the grey-scale array resulting from the trial block is subtracted from the grey-scale array resulting from the reference block, generating a 128×128 array of error values. This error data is used to tweak the arrangement of lines in the trial block.

[0300] In cells of the trial calibration block where the error value is positive, the line is too long. That is, the pattern is too dark in those cells (i.e. it has a low luminance grey-scale value), due to a surplus of line length (i.e. too much ink). By shortening the line length in those cells, their luminance is increased (i.e. the cell is lightened). Shortening can be effected by straightening curved arcs, or by relocating a line's entrance and exit points in a cell so less distance is traversed through the cell.

[0301] Conversely, in cells where the error value is negative, the line is too short. By increasing the line length in such cells, their luminance is decreased (i.e. the cell is darkened). Increasing the line length through a cell can be accomplished by increasing the curvature of the line in the cell, or by relocating a line's entrance and exit points along the boundary of the cell, so more distance is traversed through the cell.

[0302] A computer program is desirably employed to effect the foregoing changes in line routing to achieve the desired darkening or lightening of each cell.

[0303] After line positions in the trial calibration block have been tweaked in this fashion, the trial block is again converted to grey-scale values, and again subtracted from the reference block. Again, an array of error values is produced. The positions of the lines are then further tweaked in accordance with the error values.

[0304] The foregoing steps of tweaking line routes in accordance with error signals, converting anew into grey-scale, and computing new error values, is repeated until the luminance of the resulting weave pattern in the trial block is arbitrarily close to the luminance of the reference block. Four of five iterations of this procedure commonly suffice to converge on a final calibration block.

[0305] (It will be noted that the initial tile pattern created by the designer is done at the tile level—8×8 cells. After the initial trial tile is created, subsequent processing proceeds at the block level (128×128 cells). A common result of the iterative design procedure is that the component tiles lose their uniformity. That is, the pattern of lines in a tile at a corner of the final calibration block will generally be slightly different than the pattern of lines in a tile near the center of the block.)

[0306] After the final calibration block pattern has been established as above, the blocks are tiled repetitively over some or all of the security document, and can serve either as a background design element, or as a more apparent element of the design. By printing this weave pattern in an ink color close to the paper substrate color, the patterning is highly unobtrusive. (If a highly contrasting ink color is used, and if the pattern extends over most or all of the security document, it may be desirable to employ a brighter luminance paper than otherwise, since the weave pattern effectively darkens the substrate.)

[0307] As noted in my application Ser. No. 08/649,419 (now U.S. Pat. No. 5,862,260), the Fourier-Mellin transform has the property that the same output pattern is produced, regardless of rotation or scaling of the input image. The invariant output pattern is shifted in one dimension proportional to image rotation, and shifted in another dimension proportional to image scaling. When an image whose F-M transform is known, is thereafter rotated and/or scaled, the degree of rotation and scaling can be determined by observing the degree of shift of the transformed F-M pattern in the two dimensions. Once the rotation and scale are known, reciprocal processing of the image can be performed to restore the image to its original orientation and scale.

[0308] In the above-described embodiment, the calibration block pattern has a known F-M transform. When a security document incorporating such a pattern is scanned (e.g. by a photocopier, a flatbed scanner, a facsimile machine, etc.), the resulting data can be F-M transformed. The known F-M pattern is then identified in the transformed data, and its two-dimensional shift indicates the scale and rotation corruption of the scanned security document data. With these parameters known, misregistration of the security document—including scale and rotation corruption—can be backed-off, and the security document data restored to proper alignment and scale. In this re-registered state, the watermark can be detected. (In alternative embodiments, the original scan data is not processed to remove the scale/rotation effects. Instead, subsequent processing proceeds with the data in its corrupted state, and takes into account the specific corruption factor(s) to nonetheless yield accurate decoding, etc.)

[0309] The just-described calibration pattern and design procedure, of course, are just exemplary, and are subject to numerous modifications. The dimensions can be varied at will. It is not essential that the cell size of the calibration tiles match that of the watermark. Nor do the cells sizes need to be integrally related to each other. Nor does the calibration pattern need to be implemented as lines; other ink patterns can alternatively be used to approximate the grey-scale reference pattern

[0310] There is no requirement that the lines snake continuously through the tiles. A line can connect to just a single edge point of a tile, resulting in a line that crosses that tile boundary, but no other. Or a line can both begin and end in a single tile, and not connect to any other.

[0311] While darker lines on a lighter background are illustrated, lighter lines on a darker background can alternatively be employed.

[0312] The iterative design procedure can employ the F-M transform (or other transform). For example, the trial block pattern can be transformed to the F-M domain, and there compared with the F-M transform of the reference block. An F-M domain error signal can thus be obtained, and the routing of the lines can be changed in accordance therewith.

[0313] Although the illustrated embodiment tweaked the cell-based grey-scales of the calibration block by changing line curvature and position, other luminance changing techniques can be employed. For example, the width of the weave lines can be locally changed, or small ink dots can be introduced into certain cell areas.

[0314] The foregoing (and following) discussions contemplate that the watermark and/or calibration pattern is printed at the same time as (indeed, sometimes as part of) the line art on the security document. In many applications it is desirable to provide the calibration pattern on the security document substrate prior to printing. The markings can be ink applied by the manufacturer, or can be embossings applied, e.g., by rollers in the paper-making process. (Such textural marking is discussed further below.) Or, the markings can be applied by the security document printer, as a preliminary printing operation, such as by offset printing. By using an ink color/density that is already closely matched to the underlying tint of the paper stock, the manufacturer of the paper can introduce less tinting during its manufacture. Such tinting will effectively be replaced by the preliminary printing of the watermark/calibration pattern on the blank paper.

[0315] Calibration signals entirely different than those detailed above can also be used. Calibration signals that are optimized to detect rotation, but not scaling, can be employed when scaling is not a serious concern. DCT and Fourier transforms provide data that is readily analyzed to determine rotation. A calibration signal can be tailored to stand out in a typically low-energy portion of the transformed spectrum (e.g. a series of fine lines at an inclined angle transforms to a usually vacant region in DCT space), and the scanned image can be transformed to the DCT/Fourier domains to examine any shift in the calibration signal (e.g. a shift in the spatial frequency representation of the inclined lines).

[0316] In some security documents, the just-described calibration weave is printed independently of the watermark encoding. In other embodiments, the weave serves as the lines whose widths, locations, etc., are modulated by the watermark data, as detailed herein and in application Ser. No. 09/074,034 (now U.S. Pat. No. 6,449,377).

[0317] In an illustrative embodiment, the printing of the security document is achieved by intaglio printing. Intaglio is a well known printing process employing a metal plate into which the security document pattern is etched or engraved. Ink is applied to the plate, filling the etched recesses/grooves. Paper is then pressed into the plate at a very high pressure (e.g. 10-20 tons), both raised-inking and slightly deforming (texturing) the paper.

[0318] Although ink is commonly used in the intaglio process, it need not be in certain embodiments of the present invention. Instead, the paper texturing provided by the intaglio pressing—alone—can suffice to convey watermark data. (Texturing of a medium to convey watermark information is disclosed in various of my prior applications, including application Ser. No. 08/438,159 (now U.S. Pat. No. 5,850,481).)

[0319] To illustrate, an intaglio plate was engraved (using a numerically controlled engraving apparatus), to a depth of slightly less than 1 mm, in accordance with a 3.2×3.2 cm. noise-like block of watermark data. The watermark data was generated as described above (e.g. 128 bits of data, randomly distributed in a 128×128 cell array), and summed with a correspondingly-sized block of calibration data (implemented as discrete grey-scaled cells, rather than the line/weave pattern detailed above). In this embodiment, the data was not kept within a small range of digital numbers, but instead was railed to a full 8-bit dynamic range.) Banknote paper was intaglio-pressed into this plate—without ink—yielding a generally flat substrate with a 3.2×3.2 cm textured region therein. Only on fairly close inspection was the texturing visible; on casual inspection the paper surface appeared uniform.

[0320] This textured paper was placed—textured extrema down—on the platen of an conventional flatbed scanner (of the sort commonly sold as an accessory for personal computers), and scanned. The resulting image data was input to Adobe's Photoshop image processing software, version 4.0, which includes Digimarc watermark reader software. The software readily detected the watermark from the textured paper, even when the paper was skewed on the scanner platen.

[0321] The optical detection process by which a seemingly blank piece of paper can reliably convey 128 bits of data through an inexpensive scanner has not been analyzed in detail; the degree of localized reflection from the paper may be a function of whether the illuminated region is concave or convex in shape. Regardless of the explanation, it is a remarkable phenomenon to witness.

[0322] A second experiment was conducted with the same engraved plate, this time using transparent ink. The results were similar, although detection of the watermark data was not always as reliable as in the inkless case. The raised transparent ink may serve as light conduit, dispersing the incident illumination in unpredictable ways as contrasted with simple reflection off un-inked paper.

[0323] Experiments have also been conducted using traditional opaque inks. Again, the watermark can reliably be read.

[0324] In addition to the just-described technique for “reading” intaglio markings by a conventional scanner, a variant technique is disclosed in Van Renesse, Optical Inspection Techniques for Security Instrumentation, SPIE Proc. Vol. 2659, pp. 159-167 (1996), and can alternatively be used in embodiments according to the present invention.

[0325] Although intaglio is a preferred technique for printing security documents, it is not the only such technique. Other familiar techniques by which watermarks and calibration patterns can be printed include offset litho and letterpress, as well as inkjet printing, xerographic printing, etc. And, as noted, textured watermarking can be effected as part of the paper-making process, e.g. by high pressure textured rollers.

[0326] In still other embodiments, the watermark and/or calibration (“information”) patterns are not printed on the security document substrate, but rather are formed on or in an auxiliary layer that is laminated with a base substrate. If a generally clear laminate is used, the information patterns can be realized with opaque inks, supplementing the design on the underlying substrate. Or the added information can be encoded in textural form. Combinations of the foregoing can similarly be used.

[0327] To retrofit existing security document designs with information patterns, the existing artwork must be modified to effect the necessary additions and/or tweaks to localized security document luminance and/or texture.

[0328] When designing new security documents, it would be advantageous to facilitate integration of information patterns into the basic design. One such arrangement is detailed in the following discussion.

[0329] Many security documents are still designed largely by hand. A designer works at a drafting table or computer workstation, and spends many hours laying-out minute (e.g. 5 mm×5 mm) excerpts of the design. To aid integration of watermark and/or calibration pattern data in this process, an accessory layout grid can be provided, identifying the watermark “bias” (e.g. −3 to +3) that is to be included in each 250 micron cell of the security document. If the accessory grid indicates that the luminance should be slightly increased in a cell (e.g. 1%), the designer can take this bias in mind when defining the composition of the cell and include a touch less ink than might otherwise be included. Similarly, if the accessory grid indicates that the luminance should be somewhat strongly increased in a cell (e.g. 5%), the designer can again bear this in mind and try to include more ink than might otherwise be included. Due to the substantial redundancy of most watermark encoding techniques, strict compliance by the designer to these guidelines is not required. Even loose compliance can result in artwork that requires little, if any, further modification to reliably convey watermark and/or calibration information.

[0330] Such “designing-in” of embedded information in security documents is facilitated by the number of arbitrary design choices made by security document designers. A few examples from U.S. banknotes include the curls in the presidents' hair, the drape of clothing, the clouds in the skies, the shrubbery in the landscaping, the bricks in the pyramid, the fill patterns in the lettering, and the great number of arbitrary guilloche patterns and other fanciful designs, etc. All include curves, folds, wrinkles, shadow effects, etc., about which the designer has wide discretion in selecting local luminance, etc. Instead of making such choices arbitrarily, the designer can make these choices deliberately so as to serve an informational—as well as an aesthetic—function.

[0331] To further aid the security document designer, data defining several different information-carrying patterns (both watermark and/or calibration pattern) can be stored on mass storage of a computer a workstation and serve as a library of design elements for future designs. The same user-interface techniques that are employed to pick colors in image-editing software (e.g. Adobe Photoshop) and fill textures in presentation programs (e.g. Microsoft PowerPoint) can similarly be used to present a palette of information patterns to a security document designer. Clicking on a visual representation of the desired pattern makes the pattern available for inclusion in a security document being designed (e.g. filling a desired area).

[0332] In the embodiment earlier-described, the calibration pattern is printed as a visible artistic element of the security document. However, the same calibration effect can be provided subliminally if desired. That is, instead of generating artwork mimicking the grey-scale pattern of the reference calibration block, the reference calibration block can itself be encoded into the security document as small changes in local luminance. In many such embodiments, the bias to localized document luminance due to the calibration pattern is simply added to the bias due to the watermark data, and encoded like the watermark data (e.g. as localized changes to the width or position of component line-art lines, as inserted ink droplets, etc.).

[0333] The uses to which the 128 bits of watermark data can be put in security documents are myriad. Many are detailed in the materials cited above. Examples include postal stamps encoded with their value, or with the zip code of the destination to which they are addressed (or from which they were sent); banknotes encoded with their denomination, and their date and place of issuance; identification documents encoded with authentication information by which a person's identify can be verified; etc., etc.

[0334] The encoded data can be in a raw form—available to any reader having the requisite key data (in watermarking techniques where a key data is used), or can be encrypted, such as with public key encryption techniques, etc. The encoded data can embody information directly, or can be a pointer or an index to a further collection of data in which the ultimate information desired is stored.

[0335] For example, watermark data in a passport need not encode a complete dossier of information on the passport owner. Instead, the encoded data can include key data (e.g. a social security number) identifying a particular record in a remote database in which biographical data pertaining to the passport owner is stored. A passport processing station employing such an arrangement is shown in FIG. 31.

[0336] To decode watermark data, the security document must be converted into electronic image data for analysis. This conversion is typically performed by a scanner.

[0337] Scanners are well known, so a detailed description is not provided here. Suffice it to say that scanners conventionally employ a line of closely spaced photodetector cells that produce signals related to the amount of the light reflected from successive swaths of the document. Most inexpensive consumer scanners have a resolution of 300 dots per inch (dpi), or a center to center spacing of component photodetectors of about 84 microns. Higher quality scanners of the sort found in most professional imaging equipment and photocopiers have resolutions of 600 dpi (42 microns), 1200 dpi (21 microns), or better.

[0338] Taking the example of a 300 dpi scanner (84 micron photodetector spacing), each 250 micron region 12 on the security document will correspond to about a 3×3 array of photodetector samples. Naturally, only in rare instances will a given region be physically registered with the scanner so that nine photodetector samples capture the luminance in that region, and nothing else. More commonly, the image is rotated with respect to the scanner photodetectors, or is longitudinally misaligned (i.e. some photodetectors image sub-parts of two adjoining regions). However, since the scanner oversamples the regions, the luminance of each region can unambiguously be determined.

[0339] In one embodiment, the scanned data from the document is collected in a two dimensional array of data and processed to detect the embedded calibration information. The scanner data is then processed to effect a virtual re-registration of the document image. A software program next analyzes the statistics of the re-registered data (using the techniques disclosed in my prior writings) to extract the bits of the embedded data.

[0340] (Again, the reference to my earlier watermark decoding techniques is exemplary only. Once scanning begins and the data is available in sampled form, it is straightforward to apply any other watermark decoding technique to extract a correspondingly-encoded watermark. Some of these other techniques employ domain transformations (e.g. to wavelet, DCT, or Fourier domains, as part of the decoding process).)

[0341] In a variant embodiment, the scanned data is not assembled in a complete array prior to processing. Instead, it is processed in real-time, as it is generated, in order to detect embedded watermark data without delay. (Depending on the parameters of the scanner, it may be necessary to scan a half-inch or so of the document before the statistics of the resulting data unambiguously indicate the presence of a watermark.)

[0342] In other embodiments, hardware devices are provided with the capability to recognize embedded watermark data in any document images they process, and to respond accordingly.

[0343] One example is a color photocopier. Such devices employ a color scanner to generate sampled (pixel) data corresponding to an input media (e.g. a dollar bill). If watermark data associated with a security document is detected, the photocopier can take one or more steps.

[0344] One option is simply to interrupt copying, and display a message reminding the operator that it is illegal to reproduce currency.

[0345] Another option is to dial a remote service and report the attempted banknote reproduction. Photocopiers with dial-out capabilities are known in the art (e.g. U.S. Pat. No. 5,305,199) and are readily adapted to this purpose. The remote service can be an independent service, or can be a government agency.

[0346] Yet another option is to permit the copying, but to insert forensic tracer data in the resultant copy. This tracer data can take various forms. Steganographically encoded binary data is one example. An example is shown in U.S. Pat. No. 5,568,268, which is hereby incorporated by reference. The tracer data can memorialize the serial number of the machine that made the copy and/or the date and time the copy was made. To address privacy concerns, such tracer data is not normally inserted in all photocopied output, but is inserted only when the subject being photocopied is detected as being a security document. (An example of such an arrangement is shown in FIG. 32.)

[0347] Desirably, the scan data is analyzed on a line-by-line basis in order to identify illicit photocopying with a minimum of delay. If a security document is scanned, one or more lines of scanner output data may be provided to the photocopier's reprographic unit before the recognition decision has been made. In this case the photocopy will have two regions: a first region that is not tracer-marked, and a second, subsequent region in which the tracer data has been inserted.

[0348] Photocopiers with other means to detect not-to-be-copied documents are known in the art, and employ various response strategies. Examples are detailed in U.S. Pat. Nos. 5,583,614, 4,723,149, 5,633,952, 5,640,467, and 5,424,807.

[0349] Another hardware device that can employ the foregoing principles is a standalone scanner. A programmed processor (or dedicated hardware) inside the scanner analyzes the data being generated by the device, and responds accordingly.

[0350] Yet another hardware device that can employ the foregoing principles is a printer. A processor inside the device analyzes graphical image data to be printed, looking for watermarks associated with security documents.

[0351] For both the scanner and printer devices, response strategies can include disabling operation, or inserting tracer information. (Such devices typically do not have dial-out capabilities.)

[0352] Again, it is desirable to process the scanner or printer data as it becomes available, so as to detect any security document processing with a minimum of delay. Again, there will be some lag time before a detection decision is made. Accordingly, the scanner or printer output will be comprised of two parts, one without the tracer data, and another with the tracer data.

[0353] Many security documents already include visible structures that can be used as aids in banknote detection (e.g. the seal of the issuing central bank, and various geometrical markings). In accordance with a further aspect of the present invention, a security document is analyzed by an integrated system that considers both the visible structures and watermark-embedded data.

[0354] Visible security document structures can be sensed using known pattern recognition techniques. Examples of such techniques are disclosed in U.S. Pat. Nos. 5,321,773, 5,390,259, 5,533,144, 5,539,841, 5,583,614, 5,633,952, 4,723,149, 5,692,073, and 5,424,807 and laid-open foreign applications EP 649,114 and EP 766,449.

[0355] In photocopiers (and the like) equipped to detect both visible structures and watermarks from security documents, the detection of either can cause one or more of the above-noted responses to be initiated (FIG. 32).

[0356] Again, scanners and printers can be equipped with a similar capability—analyzing the data for either of these security document hallmarks. If either is detected, the software (or hardware) responds accordingly.

[0357] Identification of security documents by watermark data provides an important advantage over recognition by visible structures—it cannot so easily be defeated. A security document can be doctored (e.g. by white-out, scissors, or less crude techniques) to remove/obliterate the visible structures. Such a document can then be freely copied on either a visible structure-sensing photocopier or scanner/printer installation. The removed visible structure can then be added back in via a second printing/photocopying operation. If the printer is not equipped with security document-disabling capabilities, image-editing tools can be used to insert visible structures back into image data sets scanned from such doctored documents, and the complete document can then be freely printed. By additionally including embedded watermark data in the security document, and sensing same, such ruses will not succeed.

[0358] (A similar ruse is to scan a security document image on a non-security document-sensing scanner. The resulting image set can then be edited by conventional image editing tools to remove/obliterate the visible structures. Such a data set can then be printed—even on a printer/photocopier that examines such data for the presence of visible structures. Again, the missing visible structures can be inserted by a subsequent printing/photocopying operation.)

[0359] Desirably, the visible structure detector and the watermark detector are integrated together as a single hardware and/or software tool. This arrangement provides various economies, e.g., in interfacing with the scanner, manipulating pixel data sets for pattern recognition and watermark extraction, electronically re-registering the image to facilitate pattern recognition/watermark extraction, issuing control signals (e.g. disabling) signals to the photocopier/scanner, etc.

[0360] While the foregoing apparatuses are particularly concerned with counterfeit deterrence, the embedded markings can also serve other functions. Examples include banknote processing machines that perform denomination sorting, counterfeit detection, and circulation analysis functions. (i.e., banknotes with certain markings may be distributed through known sources, and their circulation/distribution can subsequently be monitored to assist in macro-economic analyses.)

[0361] From the foregoing, it will be recognized that various embodiments according to the present invention provide techniques for embedding multi-bit binary data in security documents, and provide for the reliable extraction of such data even in the presence of various forms of corruption (e.g. scale and rotation).

[0362] Having described and illustrated the principles of my invention with reference to several illustrative embodiments, it will be recognized that these embodiments are exemplary only and should not be taken as limiting the scope of my invention. Guided by the foregoing teachings, it should be apparent that other watermarking, decoding, and anti-counterfeiting technologies can be substituted for, and/or combined with, the elements detailed above to yield advantageous effects. Other features disclosed in my earlier applications can similarly be employed in embodiments of the technology detailed herein. (Thus, I have not here belabored application of each of the techniques disclosed in my earlier applications—e.g. use of neural networks for watermark detectors—to the present subject matter since same is fairly taught by reading the present disclosure in the context of my earlier work.)

[0363] While the technology has been described with reference to embodiments employing regular rectangular arrays of cells, those skilled in the art will recognize that other arrays—neither rectangular nor regular—can alternatively be used.

[0364] While the embodiments have described the calibration patterns as adjuncts to digital watermarks—facilitating their detection, such patterns have utility apart from digital watermarks. One example is in re-registering scanned security document image data to facilitate detection of visible structures (e.g. detection of the seal of the issuing central bank, using known pattern recognition techniques). Indeed, the use of such calibration patterns to register both watermark and visible structure image data for recognition is an important economy that can be gained by integration a visible structure detector and a watermark detector into a single system.

[0365] Although security documents have most commonly been printed (e.g. cotton/linen), other substrates are gaining in popularity (e.g. synthetics, such as polymers) and are well (or better) suited for use with the above-described techniques.

[0366] The embodiments detailed above can be implemented in dedicated hardware (e.g. ASICs), programmable hardware, and/or software, including drivers, and specifically printer and scanner drivers, as set forth in Ser. No. 09/465,418, incorporated by reference above.

Specification of 60/082,228

[0367] Watermarking is a quickly growing field of endeavor, with several different approaches. The present assignee's work is reflected in U.S. Pat. Nos. 5,710,834, 5,636,292, 5,721,788, U.S. application Ser. No. 08/327,426 (now U.S. Pat. No. 5,768,246), Ser. Nos. 08/598,083, 08/436,134 (now U.S. Pat. No. 5,748,763), Ser. No. 08/436,102 (now U.S. Pat. No. 5,748,783), and Ser. No. 08/614,521 (now U.S. Pat. No. 5,745,604), and laid-open PCT application WO97/43736. Other work is illustrated by U.S. Pat. Nos. 5,734,752, 5,646,997, 5,659,726, 5,664,018, 5,671,277, 5,687,191, 5,687,236, 5,689,587, 5,568,570, 5,572,247, 5,574,962, 5,579,124, 5,581,500, 5,613,004, 5,629,770, 5,461,426, 5,743,631, 5,488,664, 5,530,759, 5,539,735, 4,943,973, 5,337,361, 5,404,160, 5,404,377, 5,315,098, 5,319,735, 5,337,362, 4,972,471, 5,161,210, 5,243,423, 5,091,966, 5,113,437, 4,939,515, 5,374,976, 4,855,827, 4,876,617, 4,939,515, 4,963,998, 4,969,041, and published foreign applications WO 98/02864, EP 822,550, WO 97/39410, WO 96/36163, GB 2,196,167, EP 777,197, EP 736,860, EP 705,025, EP 766,468, EP 782,322, WO 95/20291, WO 96/26494, WO 96/36935, WO 96/42151, WO 97/22206, WO 97/26733. Some of the foregoing patents relate to visible watermarking techniques. Other visible watermarking techniques (e.g. data glyphs) are described in U.S. Pat. Nos. 5,706,364, 5,689,620, 5,684,885, 5,680,223, 5,668,636, 5,640,647, 5,594,809.

[0368] Most of the work in watermarking, however, is not in the patent literature but rather in published research. In addition to the patentees of the foregoing patents, some of the other workers in this field (whose watermark-related writings can by found by an author search in the INSPEC database) include I. Pitas, Eckhard Koch, Jian Zhao, Norishige Morimoto, Laurence Boney, Kineo Matsui, A. Z. Tirkel, Fred Mintzer, B. Macq, Ahmed H. Tewfik, Frederic Jordan, Naohisa Komatsu, and Lawrence O'Gorman.

[0369] The artisan is assumed to be familiar with the foregoing prior art.

[0370] In the following disclosure it should be understood that references to watermarking encompass not only the assignee's watermarking technology, but can likewise be practiced with any other watermarking technology, such as those indicated above.

[0371] Watermarking can be applied to myriad forms of information. These include imagery (including video) and audio—whether represented in digital form (e.g. an image comprised of pixels, digital video, etc.), or in an analog representation (e.g. non-sampled music, printed imagery, banknotes, etc.) Watermarking can be applied to digital content (e.g. imagery, audio) either before or after compression. Watermarking can also be used in various “description” or “synthesis” language representations of content, such as Structured Audio, Csound, NetSound, SNHC Audio and the like (c.f. http://sound.media.mit.edu/mpeg4/) by specifying synthesis commands that generate watermark data as well as the intended audio signal. Watermarking can also be applied to ordinary media, whether or not it conveys information. Examples include paper, plastics, laminates, paper/film emulsions, etc. A watermark can embed a single bit of information, or any number of bits.

[0372] The physical manifestation of watermarked information most commonly takes the form of altered signal values, such as slightly changed pixel values, picture luminance, picture colors, DCT coefficients, instantaneous audio amplitudes, etc. However, a watermark can also be manifested in other ways, such as changes in the surface microtopology of a medium, localized chemical changes (e.g. in photographic emulsions), localized variations in optical density, localized changes in luminescence, etc. Watermarks can also be optically implemented in holograms and conventional paper watermarks.

[0373] One improvement to existing technology is to employ established web crawler services (e.g. AltaVista, Excite, or Inktomi) to search for watermarked content (on the Web, in internet news groups, BBS systems, on-line systems, etc.) in addition to their usual data collecting/indexing operations. Such crawlers can download files that may have embedded watermarks (e.g. *.JPG, *.WAV, etc.) for later analysis. These files can be processed, as described below, in real time. More commonly, such files are queued and processed by a computer distinct from the crawler computer. Instead of performing watermark-read operations on each such file, a screening technique can be employed to identify those most likely to be conveying watermark data. One such technique is to perform a DCT operation on an image, and look for spectral coefficients associated with certain watermarking techniques (e.g. coefficients associated with an inclined embedded subliminal grid). To decode spread-spectrum based watermarks, the analyzing computer requires access to the noise signal used to spread the data signal. In one embodiment, interested parties submit their noise/key signals to the crawler service so as to enable their marked content to be located. The crawler service maintains such information in confidence, and uses different noise signals in decoding an image (image is used herein as a convenient shorthand for imagery, video, and audio) until watermarked data is found (if present). This allows the use of web crawlers to locate content with privately-coded watermarks, instead of just publicly-coded watermarks as is presently the case. The queueing of content data for analysis provides certain opportunities for computational shortcuts. For example, like-sized images (e.g. 256×256 pixels) can be tiled into a larger image, and examined as a unit for the presence of watermark data. If the decoding technique (or the optional pre-screening technique) employs a DCT transform or the like, the block size of the transform can be tailored to correspond to the tile size (or some integral fraction thereof). Blocks indicated as likely having watermarks can then be subjected to a full read operation. If the queued data is sorted by file name, file size, or checksum, duplicate files can be identified. Once such duplicates are identified, the analysis computer need consider only one instance of the file. If watermark data is decoded from such a file, the content provider can be informed of each URL at which copies of the file were found.

[0374] Some commentators have observed that web crawler-based searches for watermarked images can be defeated by breaking a watermarked image into sub-blocks (tiles). HTML instructions, or the like, cause the sub-blocks to be presented in tiled fashion, recreating the complete image. However, due to the small size of the component sub-blocks, watermark reading is not reliably accomplished.

[0375] This attack is overcome by instructing the web-crawler to collect the display instructions (e.g. HTML) by which image files are positioned for display on a web page, in addition to the image files themselves. Before files collected from a web page are scrutinized for watermarks, they can be concatenated in the arrangement specified by the display instructions. By this arrangement, the tiles are reassembled, and the watermark data can be reliably recovered.

[0376] Another such postulated attack against web crawler detection of image watermarks is to scramble the image (and thus the watermark) in a file, and employ a Java applet or the like to unscramble the image prior to viewing. Existing web crawlers inspect the file as they find it, so the watermark is not detected. However, just as the Java descrambling applet can be invoked when a user wishes access to a file, the same applet can similarly be employed in a web crawler to overcome such attempted circumvention of watermark detection.

[0377] Although “content” can be located and indexed by various web crawlers, the contents of the “content” are unknown. A *.JPG file, for example, may include pornography, a photo of a sunset, etc.

[0378] Watermarks can be used to indelibly associate meta-data within content (as opposed to stored in a data structure that forms another part of the object, as is conventionally done with meta-data). The watermark can include text saying “sunset” or the like. More compact information representations can alternatively be employed (e.g. coded references). Still further, the watermark can include (or consist entirely of) a Unique ID (UID) that serves as an index (key) into a network-connected remote database containing the meta data descriptors. By such arrangements, web crawlers and the like can extract and index the meta-data descriptor tags, allowing searches to be conducted based on semantic descriptions of the file contents, rather than just by file name.

[0379] Existing watermarks commonly embed information serving to communicate copyright information. Some systems embed text identifying the copyright holder. Others embed a UID which is used as an index into a database where the name of the copyright owner, and associated information, is stored.

[0380] Looking ahead, watermarks should serve more than as silent copyright notices. One option is to use watermarks to embed “intelligence” in content. One form of intelligence is knowing its “home.” “Home” can be the URL of a site with which the content is associated. A photograph of a car, for example, can be watermarked with data identifying the web site of an auto-dealer that published the image. Wherever the image goes, it serves as a link back to the original disseminator. The same technique can be applied to corporate logos. Wherever they are copied on the internet, a suitably-equipped browser or the like can decode the data and link back to the corporation's home page. (Decoding may be effected by positioning the cursor over the logo and pressing the right-mouse button, which opens a window of options—one of which is Decode Watermark.)

[0381] To reduce the data load of the watermark, the intelligence need not be wholly encoded in the content's watermark. Instead, the watermark can again provide a UID—this time identifying a remote database record where the URL of the car dealer, etc., can be retrieved. In this manner, images and the like become marketing agents—linking consumers with vendors (with some visual salesmanship thrown in). In contrast to the copyright paradigm, in which dissemination of imagery was an evil sought to be tracked and stopped, dissemination of the imagery can now be treated as a selling opportunity. A watermarked image becomes a portal to a commercial transaction.

[0382] (Using an intermediate database between a watermarked content file and its ultimate home (i.e. indirect linking) serves an important advantage: it allows the disseminator to change the “home” simply by updating a record in the database. Thus, for example, if one company is acquired by another, the former company's smart images can be made to point to the new company's home web page by updating a database record. In contrast, if the old company's home URL is hard-coded (i.e. watermarked) in the object, it may point to a URL that eventually is abandoned. In this sense, the intermediate database serves as a switchboard that couples the file to its current home.

[0383] The foregoing techniques are not limited to digital content files. The same approach is equally applicable with printed imagery, etc. A printed catalog, for example, can include a picture illustrating a jacket. Embedded in the picture is watermarked data. This data can be extracted by a simple hand-scanner/decoder device using straightforward scanning and decoding techniques (e.g. those known to artisans in those fields). In watermark-reading applications employing hand-scanners and the like, it is important that the watermark decoder be robust to rotation of the image, since the catalog photo will likely be scanned off-axis. One option is to encode subliminal graticules (e.g. visualization synchronization codes) in the catalog photo so that the set of image data can be post-processed to restore it to proper alignment prior to decoding.

[0384] The scanner/decoder device can be coupled to a modem-equipped computer, a telephone, or any other communications device. In the former instance, the device provides URL data to the computer's web browser, linking the browser to the catalog vendor's order page. (The device need not include its own watermark decoder; this task can be performed by the computer.) The vendor's order page can detail the size and color options of the jacket, inventory availability, and solicit ordering instructions (credit card number, delivery options, etc.)—as is conventionally done with on-line merchants. Such a device connected to a telephone can dial the catalog vendor's toll-free automated order-taking telephone number (known, e.g., from data encoded in the watermark), and identify the jacket to the order center. Voice prompts can then solicit the customer's choice of size, color, and delivery options, which are input by Touch Tone instructions, or by voiced words (using known voice recognition software at the vendor facility).

[0385] In such applications, the watermark may be conceptualized as an invisible bar code employed in a purchase transaction. Here, as elsewhere, the watermark can serve as a seamless interface bridging the print and digital worlds

[0386] Another way of providing content with intelligence is to use the watermark to provide Java or ActiveX code. The code can be embedded in the content, or can be stored remotely and linked to the content. When the watermarked object is activated, the code can be executed (either automatically, or at the option of the user). This code can perform virtually any function. One is to “phone home”—initiating a browser and linking to the object's home. The object can then relay any manner of data to its home. This data can specify some attribute of the data, or its use. The code can also prevent accessing the underlying content until permission is received. An example is a digital movie that, when double-clicked, automatically executes a watermark-embedded Java applet which links through a browser to the movie's distributor. The user is then prompted to input a credit card number. After the number has been verified and a charge made, the applet releases the content of the file to the computer's viewer for viewing of the movie. Support for these operations is desirably provided via the computer's operating system, or plug-in software.

[0387] Such arrangements can also be used to collect user-provided demographic information when smart image content is accessed by the consumer of the content. The demographic information can be written to a remote database and can be used for market research, customization of information about the content provided to the consumer, sales opportunities, advertising, etc.

[0388] In audio and video and the like, watermarks can serve to convey related information, such as links to WWW fan sites, actor biographies, advertising for marketing tie-ins (T-shirts, CDs, concert tickets). In such applications, it is desirable (but not necessary) to display on the user interface (e.g. screen) a small logo to signal the presence of additional information. When the consumer selects the logo via some selection device (mouse, remote control button, etc.), the information is revealed to the consumer, who can then interact with it.

[0389] Much has been written (and patented) on the topic of asset rights management. Sample patent documents include U.S. Pat. Nos. 5,715,403, 5,638,443, 5,634,012, 5,629,980. Again, much of the technical work is memorialized in journal articles, which can be identified by searching for relevant company names and trademarks such as IBM's Cryptolope system, Portland Software's ZipLock system, the Rights Exchange service by Softbank Net Solutions, and the DigiBox system from InterTrust Technologies.

[0390] An exemplary asset management system makes content available (e.g. from a web server, or on a new computer's hard disk) in encrypted form. Associated with the encrypted content is data identifying the content (e.g. a preview) and data specifying various rights associated with the content. If a user wants to make fuller use of the content, the user provides a charge authorization (e.g. a credit card) to the distributor, who then provides a decryption key, allowing access to the content. (Such systems are often realized using object-based technology. In such systems, the content is commonly said to be distributed in a “secure container.”)

[0391] Desirably, the content should be marked (personalized/serialized) so that any illicit use of the content (after decryption) can be tracked. This marking can be performed with watermarking, which assures that the mark travels with the content wherever—and in whatever form—it may go. The watermarking can be effected by the distributor—prior to dissemination of the encrypted object—such as by encoding a UID that is associated in a database with that particular container. When access rights are granted to that container, the database record can be updated to reflect the purchaser, the purchase date, the rights granted, etc. An alternative is to include a watermark encoder in the software tool used to access (e.g. decrypt) the content. Such an encoder can embed watermark data in the content as it is released from the secure container, before it is provided to the user. The embedded data can include a UID, as described above. This UID can be assigned by the distributor prior to disseminating the container. Alternatively, the UID can be a data string not known or created until access rights have been granted. In addition to the UID, the watermark can include other data not known to the distributor, e.g. information specific to the time(s) and manner(s) of accessing the content.

[0392] In other systems, access rights systems can be realized with watermarks without containers etc. Full resolution images, for example, can be freely available on the web. If a user wishes to incorporate the imagery into a web page or a magazine, the user can interrogate the imagery as to its terms and conditions of use. This may entail linking to a web site specified by the embedded watermark (directly, or through an intermediate database), which specifies the desired information. The user can then arrange the necessary payment, and use the image knowing that the necessary rights have been secured.

[0393] As noted, digital watermarks can also be realized using conventional (e.g. paper) watermarking technologies. Known techniques for watermarking media (e.g. paper, plastic, polymer) are disclosed in U.S. Pat. Nos. 5,536,468, 5,275,870, 4,760,239, 4,256,652, 4,370,200, and 3,985,927 and can be adapted to display of a visual watermark instead of a logo or the like. Note that some forms of traditional watermarks which are designed to be viewed with transmissive light can also show up as low level signals in reflective light, as is typically used in scanners. Transmissive illumination detection systems can also be employed to detect such watermarks, using optoelectronic traditional-watermark detection technologies known in the art.

[0394] As also noted, digital watermarks can be realized as part of optical holograms. Known techniques for producing and securely mounting holograms are disclosed in U.S. Pat. Nos. 5,319,475, 5,694,229, 5,492,370, 5,483,363, 5,658,411 and 5,310,222. To watermark a hologram, the watermark can be represented in the image or data model from which the holographic diffraction grating is produced. In one embodiment, the hologram is produced as before, and displays an object or symbol. The watermark markings appear in the background of the image so that they can be detected from all viewing angles. In this context, it is not critical that the watermark representation be essentially imperceptible to the viewer. If desired, a fairly visible noise-like pattern can be used without impairing the use to which the hologram is put.

[0395] Digital watermarks can also be employed in conjunction with labels and tags. In addition to conventional label/tag printing processes, other techniques—tailored to security—can also be employed. Known techniques useful in producing security labels/tags are disclosed in U.S. Pat. Nos. 5,665,194, 5,732,979, 5,651,615, and 4,268,983. The imperceptibility of watermarked data, and the ease of machine decoding, are some of the benefits associated with watermarked tags/labels. Additionally, the cost is far less than many related technologies (e.g. holograms). Watermarks in this application can be used to authenticate the originality of a product label, either to the merchant or to the consumer of the associated product, using a simple scanner device, thereby reducing the rate of counterfeit product sales.

[0396] Recent advances in color printing technology have greatly increased the level of casual counterfeiting. High quality scanners are now readily available to many computer users, with 300 dpi scanners available for under $100, and 600 dpi scanners available for marginally more. Similarly, photographic quality color ink-jet printers are commonly available from Hewlett-Packard Co., Epson, etc. for under $300.

[0397] Watermarks in banknotes and other security documents (passports, stock certificates, checks, etc.—all collectively referred to as banknotes herein) offer great promise to reduce such counterfeiting, as discussed more fully below. Additionally, watermarks provide a high-confidence technique for banknote authentication. One product enabled by this increased confidence is automatic teller machines that accept, as well as dispense, cash. The machine is provided with known optical scanning technology to produce digital data corresponding to the face(s) of the bill. This image set is then analyzed to extract the watermark data. In watermarking technologies that require knowledge of a code signal for decoding (e.g. noise modulation signal, crypto key, spreading signal, etc.), a bill may be watermarked in accordance with several such codes. Some of these codes are public—permitting their reading by conventional machines. Others are private, and are reserved for use by government agencies and the like. (C.f. public and private codes in the present assignee's issued patents.)

[0398] Banknotes presently include certain markings which can be used as an aid in note authentication. Well known visible structures are added to banknotes to facilitate visual authentication and machine detection. An example is the seal of the issuing central bank. Others are geometrical markings. Desirably, a note is examined by an integrated detection system, for both such visible structures as well as the present watermark-embedded data, to determine authenticity.

[0399] The visible structures can be sensed using known pattern recognition techniques. Examples of such techniques are disclosed in U.S. Pat. Nos. 5,321,773, 5,390,259, 5,533,144, 5,539,841, 5,583,614, 5,633,952, 4,723,149 and 5,424,807 and laid-open foreign application EP 766,449. The embedded watermark data can be recovered using the scanning/analysis techniques disclosed in the cited patents and publications.

[0400] To reduce counterfeiting, it is desirable that document-reproducing technologies recognize banknotes and refuse to reproduce same. A photocopier, for example, can sense the presence of either a visible structure *or* embedded banknote watermark data, and disable copying if either is present. Scanners and printers can be equipped with a similar capability—analyzing the data scanned or to be printed for either of these banknote hallmarks. If either is detected, the software (or hardware) disables further operation.

[0401] The watermark detection criteria provides an important advantage not otherwise available. An original bill can be doctored (e.g. by white-out, scissors, or less crude techniques) to remove/obliterate the visible structures. Such a document can then be freely copied on either a visible structure-sensing photocopier or scanner/printer installation. The removed visible structure can then be added in via a second printing/photocopying operation. If the printer is not equipped with banknote-disabling capabilities, image-editing tools can be used to insert visible structures back into image data sets scanned from such doctored bills, and the complete bill freely printed. By additionally including embedded watermark data in the banknote, and sensing same, such ruses will not succeed.

[0402] (A similar ruse is to scan a banknote image on a non-banknote-sensing scanner. The resulting image set can then be edited by conventional image editing tools to remove/obliterate the visible structures. Such a data set can then be printed—even on a printer/photocopier that examines such data for the presence of visible structures. Again, the missing visible structures can be inserted by a subsequent printing/photocopying operation.)

[0403] Desirably, the visible structure detector and the watermark detector are integrated together as a single hardware and/or software tool. This arrangement provides various economies, e.g., in interfacing with the scanner, manipulating pixel data sets for pattern recognition and watermark extraction, electronically re-registering the image to facilitate pattern recognition/watermark extraction, issuing control signals (e.g. disabling) signals to the photocopier/scanner, etc.

[0404] A related principle is to insert an imperceptible watermark having a UID into all documents printed with a printer, scanned with a scanner, or reproduced by a photocopier. The UID is associated with the particular printer/photocopier/scanner in a registry database maintained by the products' manufacturers. The manufacturer can also enter in this database the name of the distributor to whom the product was initially shipped. Still further, the owner's name and address can be added to the database when the machine is registered for warranty service. While not preventing use of such machines in counterfeiting, the embedded UID facilitates identifying the machine that generated a counterfeit banknote. (This is an application in which a private watermark might best be used.)

[0405] While the foregoing applications disabled potential counterfeiting operations upon the detection of *either* a visible structure or watermarked data, in other applications, both criteria must be met before a banknote is recognized as genuine. Such applications typically involve the receipt or acceptance of banknotes, e.g. by ATMs as discussed above.

[0406] The foregoing principles (employing just watermark data, or in conjunction with visible indicia) can likewise be used to prevent counterfeiting of tags and labels (e.g. the fake labels and tags commonly used in pirating Levis brand jeans, Microsoft software, etc.)

[0407] The reader may first assume that banknote watermarking is effected by slight alterations to the ink color/density/distribution, etc. on the paper. This is one approach. Another is to watermark the underlying medium (whether paper, polymer, etc.) with a watermark. This can be done by changing the microtopology of the medium (a la mini-Braille) to manifest the watermark data. Another option is to employ a laminate on or within the banknote, where the laminate has the watermarking manifested thereon/therein. The laminate can be textured (as above), or its optical transmissivity can vary in accordance with a noise-like pattern that is the watermark, or a chemical property can similarly vary.

[0408] Another option is to print at least part of a watermark using photoluminescent ink. This allows, e.g., a merchant presented with a banknote, to quickly verify the presence of *some* watermark-like indicia in/on the bill even without resort to a scanner and computer analysis (e.g. by examining under a black light). Such photoluminescent ink can also print human-readable indicia on the bill, such as the denomination of a banknote. (Since ink-jet printers and other common mass-printing technologies employ cyan/magenta/yellow/black to form colors, they can produce only a limited spectrum of colors. Photoluminescent colors are outside their capabilities. Fluorescent colors—such as the yellow, pink and green dyes used in highlighting markers—can similarly be used and have the advantage of being visible without a black light.)

[0409] An improvement to existing encoding techniques is to add an iterative assessment of the robustness of the mark, with a corresponding adjustment in a re-watermarking operation. Especially when encoding multiple bit watermarks, the characteristics of the underlying content may result in some bits being more robustly (e.g. strongly) encoded than others. In an illustrative technique employing this improvement, a watermark is first embedded in an object. Next, a trial decoding operation is performed. A confidence measure (e.g. signal-to-noise ratio) associated with each bit detected in the decoding operation is then assessed. The bits that appear weakly encoded are identified, and corresponding changes are made to the watermarking parameters to bring up the relative strengths of these bits. The object is then watermarked anew, with the changed parameters. This process can be repeated, as needed, until all of the bits comprising the encoded data are approximately equally detectable from the encoded object, or meet some predetermined signal-to-noise ratio threshold.

[0410] The foregoing applications, and others, can generally benefit by multiple watermarks. For example, an object (physical or data) can be marked once in the spatial domain, and a second time in the spatial frequency domain. (It should be understood that any change in one domain has repercussions in the other. Here we reference the domain in which the change is directly effected.)

[0411] Another option is to mark an object with watermarks of two different levels of robustness, or strength. The more robust watermark withstands various types of corruption, and is detectable in the object even after multiple generations of intervening distortion. The less robust watermark can be made frail enough to fail with the first distortion of the object. In a banknote, for example, the less robust watermark serves as an authentication mark. Any scanning and reprinting operation will cause it to become unreadable. Both the robust and the frail watermarks should be present in an authentic banknote; only the former watermark will be present in a counterfeit.

[0412] Still another form of multiple-watermarking is with content that is compressed. The content can be watermarked once (or more) in an uncompressed state. Then, after compression, a further watermark (or watermarks) can be applied.

[0413] Still another advantage from multiple watermarks is protection against sleuthing. If one of the watermarks is found and cracked, the other watermark(s) will still be present and serve to identify the object.

[0414] The foregoing discussion has addressed various technological fixes to many different problems. Exemplary solutions have been detailed above. Others will be apparent to the artisan by applying common knowledge to extrapolate from the solutions provided above.

[0415] For example, the technology and solutions disclosed herein have made use of elements and techniques known from the cited references. Other elements and techniques from the cited references can similarly be combined to yield further implementations within the scope of the present invention. Thus, for example, holograms with watermark data can be employed in banknotes, single-bit watermarking can commonly be substituted for multi-bit watermarking, technology described as using imperceptible watermarks can alternatively be practiced using visible watermarks (glyphs, etc.), techniques described as applied to images can likewise be applied to video and audio, local scaling of watermark energy can be provided to enhance watermark signal-to-noise ratio without increasing human perceptibility, various filtering operations can be employed to serve the functions explained in the prior art, watermarks can include subliminal graticules to aid in image re-registration, encoding may proceed at the granularity of a single pixel (or DCT coefficient), or may similarly treat adjoining groups of pixels (or DCT coefficients), the encoding can be optimized to withstand expected forms of content corruption. Thus, the exemplary embodiments are only selected samples of the solutions available by combining the teachings referenced above. The other solutions necessarily are not exhaustively described herein, but are fairly within the understanding of an artisan given the foregoing disclosure and familiarity with the cited art.

[0416] Concluding Remarks

[0417] Having described and illustrated the principles of the technology with reference to specific implementations, it will be recognized that the technology can be implemented in many other, different, forms. The techniques for embedding and detecting a watermark may be applied to various types of watermarks, including those encoded using linear or non-linear functions to apply a watermark message to a host signal. As one example, embedding methods, such as methods for error correction coding, methods for mapping watermark messages to the host signal, and methods for redundantly encoding watermark messages apply whether the watermarking functions are linear or non-linear. In addition, the techniques for determining and refining a watermark's orientation apply to linear and non-linear watermark methods. For example, the methods described above for detecting orientation of a watermark signal in a potentially transformed version of the watermarked signal apply to watermark systems that use different methods for embedding and reading messages, including, but not limited to, techniques that modulate spatial or temporal domain intensity values, that modulate transform coefficients, that employ dither modulation or quantization index modulation.

[0418] Some of the detector methods described above invoke a watermark message reader to assess the merits of a given orientation of a watermark signal in a potentially transformed version of the watermarked signal. In particular, some of these techniques assess the merits of an orientation by invoking a reader to determine the extent to which known message bits agree with those read from the watermarked signal using the orientation. These techniques are not specific to the type of message encoding or reading as noted in the previous paragraph. The merits of a given estimate of a watermark signal's orientation may be assessed by selecting an orientation parameter that increases correlation between the watermark signal (or known watermark signal attributes) and the watermarked signal, or that improves recovery of known watermark message bits from the watermark signal.

[0419] Some watermark readers extract a message from a watermarked signal by correlating known attributes of a message symbol with the watermarked signal. For example, one symbol might be associated with a first pseudorandom noise pattern, while another symbol is associated with another pseudorandom noise pattern. If the reader determines that a strong correlation between the known attribute and the watermark signal exists, then it is likely that the watermarked signal contains the message symbol.

[0420] Other watermark readers analyze the watermarked signal to identify attributes that are associated with a message symbol. Generally speaking, these watermark readers are using a form of correlation, but in a different form. If the reader identifies evidence of watermark signal attributes associated with a message symbol, it notes that the associated message symbol has likely been encoded. For example, readers that employ quantization index modulation analyze the watermarked signal by applying quantizers to the signal to determine which quantizer was most likely used in the embedder to encode a message. Since message symbols are associated with quantizers, the reader extracts a message by estimating the quantizer used to encode the message. In these schemes, the signal attribute associated with a message symbol is the type of quantization applied to the signal. Regardless of the signal attributes used to encode and read a watermark message, the methods described above for determining watermark orientation and refining orientation parameters still apply.

[0421] The paper entitled “Smart Images” Using Digimarc's Watermarking Technology, by Adnan M. Alattar, is also incorporated by reference. For additional information about a detector optimization that looks for a watermark in portions of a signal that are more likely to contain a recoverable watermark signal, see U.S. patent application Ser. No. 09/302,663, filed Apr. 30, 1999, entitled Watermark Detection Utilizing Regions with Higher Probability of Success, by Ammon Gustafson, Geoffrey Rhoads, Adnan Alattar, Ravi Sharma and Clay Davidson (now U.S. Pat. No. 6,442,284), which is hereby incorporated by reference.

[0422] To provide a comprehensive disclosure without unduly lengthening the specification, applicant incorporates by reference the patents and applications cited above. The particular combinations of elements and features in the above-detailed embodiments are exemplary only; the interchanging and substitution of these teachings with other teachings in this and the incorporated-by-reference patents/applications are also contemplated.