Extended color processing on pelican array cameras

[0001] The current application claims priority to U.S. Application No. 61/798,602 entitled “Extended Color Processing on Pelican Array Cameras,” filed Mar. 15, 2013, the disclosure of which is incorporated herein by reference.

[0002] The present invention generally relates to imaging and more specifically relates to array cameras that utilize cross-channel fusion and extended color processing.

[0003] In a typical imaging device, light enters through an opening (aperture) at one end of the imaging device and is directed to an image sensor by one or more optical elements such as lenses. The image sensor consists of pixels that generate signals upon receiving light via the optical element. Commonly used image sensors include charge-coupled device image sensors (CCDs) and complementary metal-oxide semiconductor (CMOS) sensors.

[0004] Conventional digital cameras typically achieve color separation by performing color separation in the optical path and using a separate image sensor for the wavelengths of light corresponding to each of the primary colors (i.e. RGB), using an image sensor with color separation and multiple signal collection capability within each pixel, or by applying filters over a single sensor so that individual pixels detect wavelengths of light corresponding to one of the primary colors. Use of filters is particularly common in cameras that have a small form factor, such as cameras incorporated in mobile phone handsets and other consumer electronics devices including but not limited to, laptop computers and televisions. A common filter that is formed on image sensors is the Bayer filter, the pattern of which includes 50% green filters, 25% red filters, and 25% blue filters. The output of an image sensor to which a Bayer filter is applied can be reconstructed as a color image using interpolation techniques.

[0005] Image sensors are subject to various performance constraints including, among others, dynamic range, signal to noise (SNR) ratio and low light sensitivity. The dynamic range is defined as the ratio of the maximum possible signal that can be captured by a pixel to the total noise signal. The SNR of a captured image is, to a great extent, a measure of image quality. In general, as more light is captured by the pixel, the higher the SNR. The light sensitivity of an image sensor is typically determined by the intensity of light incident upon the sensor pixels. At low light levels, each pixel's light gathering capability is constrained by the low signal levels incident upon each pixel.

[0006] A challenge associated with increasing the number of pixels in an image sensor is that the lens system is dimensioned to span the image sensor. The problem is most acute with mobile cameras, such as those used in mobile phones and consumer electronics devices, where the form factor of the lens system can significantly impact the overall form factor of the mobile device.

[0007] In response to the constraints placed upon a traditional digital camera based upon the camera obscura, a new class of cameras that can be referred to as array cameras have been proposed. Array cameras are characterized in that they include multiple arrays of pixels, each having a separate lens system. Examples of 2, 3 and 4 array cameras in which each array of pixels captures light from a different band of the visible spectrum and the captured images are combined to create a full color image is disclosed in U.S. Pat. No. 7,199,348 to Olsen et al., the disclosure of which is incorporated by reference herein in its entirety. U.S. Pat. No. 7,262,799 to Suda, the disclosure of which is incorporated herein by reference in its entirety, discloses a 2×2 array camera including one sensor used to sense a red (R) image signal, one sensor used to sense a blue (B) image signal and, two sensors used to sense green (G) image signals.

[0008] Systems and methods for extended color processing on Pelican array cameras in accordance with embodiments of the invention are disclosed. In one embodiment, a method of generating a high resolution image of a scene using an imager array including a plurality of imagers that each capture an image of the scene, and a forward imaging transformation for each imager includes obtaining input images captured by a plurality of imagers, where a first set of input images includes image information captured in a first band of visible wavelengths and a second set of input images includes image information captured in a second band of visible wavelengths and non-visible wavelengths, determining an initial estimate of at least a portion of a high resolution image using a processor configured by software to combine image information from the first set of input images into a first fused image, combine image information from the second set of input images into a second fused image, spatially register the first fused image and the second fused image, denoise the first fused image using a first bilateral filter, denoise the second fused image using a second bilateral filter, normalize the second fused image in the photometric reference space of the first fused image, and combine the first fused image and the second fused image into an initial estimate of at least a portion of the high resolution image, determining a high resolution image that when mapped through the forward imaging transformation matches the input images to within at least one predetermined criterion using the initial estimate of at least a portion of the high resolution image using the processor configured by software, where each forward imaging transformation corresponds to the manner in which each imager in the imaging array generated the input images, and where the high resolution image has a resolution that is greater than any of the input images.

[0009] In a further embodiment, the first band of visible wavelengths and the second band of visible and non-visible wavelengths have some degree of overlap.

[0010] In another embodiment, the second band of visible and non-visible wavelengths includes green, red, and near-infrared light.

[0011] In a still further embodiment, the first fused image and the second fused image have the same resolution and the resolution is higher than the resolution of any of the input images.

[0012] In still another embodiment, the first set of input images are captured by a first set of imagers from the plurality of imagers and the first set of imagers are sensitive to light in the first band of visible wavelengths, and the second set of input images are captured by a second set of imagers from the plurality of imagers and the second set of imagers are sensitive to light in the second band of visible and non-visible wavelengths.

[0013] In a yet further embodiment, the processor being configured to combine image information from the first set of input images into a first fused image utilizes analog gain and noise information from the first set of imagers and the processor being configured to combine image information from the second set of input images into a second fused image utilizes analog gain and noise information from the second set of imagers.

[0014] In yet another embodiment, the first bilateral filter and the second bilateral filter utilize weights that are a function of both the photometric and geometric distance between a pixel and pixels in the neighborhood of the pixel.

[0015] In a further embodiment again, the first bilateral filter is a cross-channel bilateral filter utilizing weights determined for the second image.

[0016] In another embodiment again, the first set of input images are captured by a first set of imagers from the plurality of imagers and the first bilateral filter is a cross-channel bilateral filter utilizing weights determined for the second image when an analog gain value of the first set of imagers is above a predetermined threshold.

[0017] In a further additional embodiment, normalizing the second fused image in the photometric reference space of the first fused image includes applying gains and offsets to pixels of the second fused image.

[0018] In another additional embodiment, the gain for each pixel of the second fused image is determined by the equation:

[0019] $\hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},$

[0020] and the bias for each pixel of the second fused image is determined by the equation:
{circumflex over (b)}=g−â·ē

[0021] where:

[0022] $\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),\stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),$

[0023] e is the second fused image, g is the first fused image, N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel, and r and c are row and column indices into the images within the bounds defined by N_r and N_c.

[0024] In a still yet further embodiment, determining an initial estimate of at least a portion of a high resolution image using a processor configured by software also includes the processor being configured to cross-channel normalize the first fused image in the photometric reference space of the second fused image.

[0025] In still yet another embodiment, the processor being configured to cross-channel normalize the first fused image in the photometric reference space of the second fused image includes the processor being configured to apply gains and offsets to pixels of the first fused image.

[0026] In a still further embodiment again, the gain for each pixel of the first fused image is determined by the equation:

[0027] $\hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},$

[0028] and the bias for each pixel of the first fused image is determined by the equation:
{circumflex over (b)}=g−â·ē

[0029] where:

[0030] $\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),\stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),$

[0031] e is the first fused image, g is the second fused image, N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel, and r and c are row and column indices into the images within the bounds defined by N_r and N_c.

[0032] In still another embodiment again, the processor being configured to normalize the second fused image in the photometric reference space of the first fused image includes the processor being configured to select a first pixel of interest in the second fused image and a first collection of similar pixels in the neighborhood of the first pixel of interest, select a second pixel of interest in the first fused image corresponding to the first pixel of interest and a second collection of similar pixels in the neighborhood of the second pixel of interest, determine the intersection of the first collection of similar pixels and the second collection of similar pixels, calculate gain and offset values using the intersection of the two collections, apply the gain and offset values to the appropriate pixels in the second fused image.

[0033] In a still further additional embodiment, the intersection of the first collection of similar pixels and the second collection of similar pixels is the set of pixels in the first and second collections having the same corresponding locations in each of the first and second fused images.

[0034] In still another additional embodiment, the gain for each pixel in the intersection of the two collections within the second fused image is determined by the equation:

[0035] $\hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},$

[0036] and the bias for each pixel in the intersection of the two collections within the second fused image is determined by the equation:
{circumflex over (b)}=g−â·ē

[0037] where:

[0038] $\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),\stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),$

[0039] e is the second fused image, g is the first fused image, N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel, and r and c are row and column indices into the images within the bounds defined by N_r and N_c.

[0040] In a yet further embodiment again, an array camera configured to generate a high resolution image of a scene using an imager array including a plurality of imagers that each capture an image of the scene, and a forward imaging transformation for each imager, includes an imager array including a plurality of imagers, and a processor configured by software to obtain input images captured by the plurality of imagers, where a first set of input images includes image information captured in a first band of visible wavelengths and a second set of input images includes image information captured in a second band of visible wavelengths and non-visible wavelengths, determine an initial estimate of at least a portion of a high resolution image by combining image information from the first set of input images into a first fused image, combining image information from the second set of input images into a second fused image, spatially registering the first fused image and the second fused image, denoising the first fused image using a first bilateral filter, denoising the second fused image using a second bilateral filter, normalizing the second fused image in the photometric reference space of the first fused image, combining the first fused image and the second fused image into an initial estimate of at least a portion of the high resolution image, determine a high resolution image that when mapped through the forward imaging transformation matches the input images to within at least one predetermined criterion using the initial estimate of at least a portion of the high resolution image using the processor configured by software, where each forward imaging transformation corresponds to the manner in which each imager in the imaging array generated the input images, and where the high resolution image has a resolution that is greater than any of the input images.

[0041] In yet another embodiment again, the first band of visible wavelengths and the second band of visible and non-visible wavelengths have some degree of overlap.

[0042] In a yet further additional embodiment, the second band of visible and non-visible wavelengths includes green, red, and near-infrared light.

[0043] In yet another additional embodiment, the first fused image and the second fused image have the same resolution and the resolution is higher than the resolution of any of the input images.

[0044] In a further additional embodiment again, the first set of input images are captured by a first set of imagers from the plurality of imagers and the first set of imagers are sensitive to light in the first band of visible wavelengths, and the second set of input images are captured by a second set of imagers from the plurality of imagers and the second set of imagers are sensitive to light in the second band of visible and non-visible wavelengths.

[0045] In another additional embodiment again, combining image information from the first set of input images into a first fused image utilizes analog gain and noise information from the first set of imagers and combining image information from the second set of input images into a second fused image utilizes analog gain and noise information from the second set of imagers.

[0046] In a still yet further embodiment again, the first bilateral filter and the second bilateral filter utilize weights that are a function of both the photometric and geometric distance between a pixel and pixels in the neighborhood of the pixel.

[0047] In still yet another embodiment again, the first bilateral filter is a cross-channel bilateral filter utilizing weights determined for the second image.

[0048] In a still yet further additional embodiment, the first set of input images are captured by a first set of imagers from the plurality of imagers and the first bilateral filter is a cross-channel bilateral filter utilizing weights determined for the second image when an analog gain value of the first set of imagers is above a predetermined threshold.

[0049] In still yet another additional embodiment, normalizing the second fused image in the photometric reference space of the first fused image includes applying gains and offsets to pixels of the second fused image.

[0050] In a yet further additional embodiment again, the gain for each pixel of the second fused image is determined by the equation:

[0051] $\hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},$

[0052] and the bias for each pixel of the second fused image is determined by the equation:
{circumflex over (b)}=g−â·ē

[0053] where:

[0054] $\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),\stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),$

[0055] e is the second fused image, g is the first fused image, N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel, and r and c are row and column indices into the images within the bounds defined by N_r and N_c.

[0056] In yet another additional embodiment again, the processor is also configured to cross-channel normalize the first fused image in the photometric reference space of the second fused image.

[0057] In a still yet further additional embodiment again, the processor being configured to cross-channel normalize the first fused image in the photometric reference space of the second fused image includes the processor being configured to apply gains and offsets to pixels of the first fused image.

[0058] In still yet another additional embodiment again, the gain for each pixel of the first fused image is determined by the equation:

[0059] $\hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},$

[0060] and the bias for each pixel of the first fused image is determined by the equation:
{circumflex over (b)}=g−â·ē

[0061] where:

[0062] $\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),\stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),$

[0063] e is the first fused image, g is the second fused image, N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel, and r and c are row and column indices into the images within the bounds defined by N_r and N_c.

[0064] In another further embodiment, normalizing the second fused image in the photometric reference space of the first fused image includes selecting a first pixel of interest in the second fused image and a first collection of similar pixels in the neighborhood of the first pixel of interest, selecting a second pixel of interest in the first fused image corresponding to the first pixel of interest and a second collection of similar pixels in the neighborhood of the second pixel of interest, determining the intersection of the first collection of similar pixels and the second collection of similar pixels, calculating gain and offset values using the intersection of the two collections, applying the gain and offset values to the appropriate pixels in the second fused image.

[0065] In still another further embodiment, the intersection of the first collection of similar pixels and the second collection of similar pixels is the set of pixels in the first and second collections having the same corresponding locations in each of the first and second fused images.

[0066] In yet another further embodiment, the gain for each pixel in the intersection of the two collections within the second fused image is determined by the equation:

[0067] $\hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},$

[0068] and the bias for each pixel in the intersection of the two collections within the second fused image is determined by the equation:
{circumflex over (b)}=g−â·ē

[0069] where:

[0070] $\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),\stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),$

[0071] e is the second fused image, g is the first fused image, N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel, and r and c are row and column indices into the images within the bounds defined by N_r and N_c.

[0086] Turning now to the drawings, systems and methods for utilizing extended color processing on image data captured by an array camera in accordance with embodiments of the invention are illustrated. Array cameras including camera modules that can be utilized to capture image data from different viewpoints (i.e. light field images) are disclosed in U.S. patent application Ser. No. 12/935,504 entitled “Capturing and Processing of Images using Monolithic Camera Array with Heterogeneous Imagers” to Venkataraman et al. In many instances, fusion and super-resolution processes such as those described in U.S. patent application Ser. No. 12/967,807 entitled “Systems and Methods for Synthesizing High Resolution Images Using Super-Resolution Processes” to Lelescu et al., can be utilized to synthesize a higher resolution (HR) 2D image or a stereo pair of higher resolution 2D images from the lower resolution (LR) images in the light field captured by an array camera. The terms high or higher resolution (HR) and low or lower resolution (LR) are used here in a relative sense and not to indicate the specific resolutions of the images captured by the array camera. The disclosures of U.S. patent application Ser. Nos. 12/935,504 and 12/967,807 are hereby incorporated by reference in their entirety.

[0087] Each two-dimensional (2D) image in a captured light field is from the viewpoint of one of the cameras in the array camera. Due to the different viewpoint of each of the cameras, parallax results in variations in the position of foreground objects within the images of the scene. Processes such as those disclosed in U.S. Provisional Patent Application No. 61/691,666 entitled “Systems and Methods for Parallax Detection and Correction in Imaged Captured Using Array Cameras” to Venkataraman et al. can be utilized to provide an accurate account of the pixel disparity as a result of parallax between the different cameras in an array. The disclosure of U.S. Patent Application Ser. No. 61/691,666 is hereby incorporated by reference in its entirety. Array cameras can use disparity between pixels in images within a light field to generate a depth map from a reference viewpoint. A depth map indicates the distance of the surfaces of scene objects from the reference viewpoint and can be utilized to determine scene dependent geometric corrections to apply to the pixels from each of the images within a captured light field to eliminate disparity when performing fusion and/or super-resolution processing.

[0088] In a number of embodiments, cross-channel fusion of a visible imaging channel and a non-visible imaging channel can be used to generate source images for a super-resolution (SR) process such as the processes described in U.S. patent application Ser. No. 12/967,807. In many embodiments, a visible channel image is fused with an extended color channel image. An extended-color channel is a channel that is designed to overlap spectrally with the visible channel spectrum. Because an extended-color channel is typically highly correlated with the visible channel with which it overlaps, fusing an extended-color image with a visible channel image can provide many advantages in SR and other image processing as will be discussed further below. Systems and methods for synthesizing images from image data captured by array cameras using cross-channel fusion and utilizing extended color processing in accordance with embodiments of the invention are discussed further below.

Array Cameras

[0089] Array cameras in accordance with embodiments of the invention can include a camera module including an array of cameras and a processor configured to read out and process image data from the camera module to synthesize images. An array camera in accordance with an embodiment of the invention is illustrated in FIG. 1. The array camera 100 includes a camera module 102 with an array of individual cameras 104 where an array of individual cameras refers to a plurality of cameras in a particular arrangement, such as (but not limited to) the square arrangement utilized in the illustrated embodiment. The camera module 102 is connected to the processor 106. The processor is also configured to communicate with one or more different types of memory 108 that can be utilized to store image data and/or contain machine readable instructions utilized to configure the processor to perform processes including (but not limited to) the various processes described below.

[0090] Processors 106 in accordance with many embodiments of the invention are configured using appropriate software to take the image data within the light field and synthesize one or more high resolution images. In several embodiments, the high resolution image is synthesized from a reference viewpoint, typically that of a reference focal plane 104 within the sensor 102. In many embodiments, the processor is able to synthesize an image from a virtual viewpoint, which does not correspond to the viewpoints of any of the focal planes 104 in the sensor 102. Unless all of the objects within a captured scene are a significant distance from the array camera, the images in the light field will include disparity due to the different fields of view of the focal planes used to capture the images. Processes for detecting and correcting for disparity are discussed further below. Although a specific array camera architecture is illustrated in FIG. 1, alternative architectures can also be utilized in accordance with embodiments of the invention.

Array Camera Modules

[0091] Array camera modules in accordance with embodiments of the invention can be constructed from an imager array or sensor including an array of focal planes and an optic array including a lens stack for each focal plane in the imager array. Sensors including multiple focal planes are discussed in U.S. patent application Ser. No. 13/106,797 entitled “Architectures for System on Chip Array Cameras”, to Pain et al., the disclosure of which is incorporated herein by reference in its entirety. Light filters can be used within each optical channel formed by the lens stacks in the optic array to enable different cameras within an array camera module to capture image data with respect to different portions of the electromagnetic spectrum.

[0092] An array camera module in accordance with an embodiment of the invention is illustrated in FIG. 1A. The array camera module 150 includes an imager array 152 including an array of focal planes 154 along with a corresponding optic array 156 including an array of lens stacks 158. Within the array of lens stacks, each lens stack 158 creates an optical channel that forms an image of the scene on an array of light sensitive pixels within a corresponding focal plane 154. Each pairing of a lens stack 158 and focal plane 154 forms a single camera 104 within the camera module. Each pixel within a focal plane 154 of a camera 104 generates image data that can be sent from the camera 104 to the processor 106. In many embodiments, the lens stack within each optical channel is configured so that pixels of each focal plane 154 sample the same object space or region within the scene. In several embodiments, the lens stacks are configured so that the pixels that sample the same object space do so with sub-pixel offsets to provide sampling diversity that can be utilized to recover increased resolution through the use of super-resolution processes.

[0093] In the illustrated embodiment, the focal planes are configured in a 5×5 array. Each focal plane 154 on the sensor is capable of capturing an image of the scene. Typically, each focal plane includes a plurality of rows of pixels that also forms a plurality of columns of pixels, and each focal plane is contained within a region of the imager that does not contain pixels from another focal plane. In many embodiments, image data capture and readout of each focal plane can be independently controlled. In this way, image capture settings including (but not limited to) the exposure times and analog gains of pixels within a focal can be determined independently to enable image capture settings to be tailored based upon factors including (but not limited to) a specific color channel and/or a specific portion of the scene dynamic range. The sensor elements utilized in the focal planes can be individual light sensing elements such as, but not limited to, traditional CIS (CMOS Image Sensor) pixels, CCD (charge-coupled device) pixels, high dynamic range sensor elements, multispectral sensor elements and/or any other structure configured to generate an electrical signal indicative of light incident on the structure. In many embodiments, the sensor elements of each focal plane have similar physical properties and receive light via the same optical channel and color filter (where present). In other embodiments, the sensor elements have different characteristics and, in many instances, the characteristics of the sensor elements are related to the color filter applied to each sensor element.

[0094] In several embodiments, color filters in individual cameras can be used to pattern the camera module with π filter groups as further discussed in U.S. Provisional Patent Application No. 61/641,165 entitled “Camera Modules Patterned with pi Filter Groups” filed May 1, 2012, the disclosure of which is incorporated by reference herein in its entirety. These cameras can be used to capture data with respect to different colors, or a specific portion of the spectrum. In contrast to applying color filters to the pixels of the camera, color filters in many embodiments of the invention are included in the lens stack. For example, a Green color camera can include a lens stack with a Green light filter that allows Green light to pass through the optical channel. A near-IR camera can include a lens stack with a near-IR light filter that allows near-IR light to pass through the optical channel.

[0095] In many embodiments, the pixels in each focal plane are the same and the light information captured by the pixels is differentiated by the color filters in the corresponding lens stack for each filter plane. Although a specific construction of a camera module with an optic array including color filters in the lens stacks is described above, camera modules including π filter groups can be implemented in a variety of ways including (but not limited to) by applying color filters to the pixels of the focal planes of the camera module similar to the manner in which color filters are applied to the pixels of a conventional color camera. In several embodiments, at least one of the cameras in the camera module can include uniform color filters applied to the pixels in its focal plane. In many embodiments, a Bayer filter pattern is applied to the pixels of one of the cameras in a camera module. In a number of embodiments, camera modules are constructed in which color filters are utilized in both the lens stacks and on the pixels of the imager array.

[0096] Although specific array cameras and imager arrays are discussed above, many different array cameras can be utilized to capture image data and synthesize images using multi-resolution depth maps as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Imager arrays in accordance with embodiments of the invention are discussed further below.

Imager Array Configurations

[0097] An imager array can include any N×M array of imagers such as the imager array (200) illustrated in FIG. 2A. In several embodiments, the array is square. In other embodiments, the array is rectangular. In a number of embodiments, the array can also be irregular. Each of the imagers typically includes its own filter and/or optical elements and can image different wavelengths of light. In a number of embodiments, the imager array includes imagers that sense red light (R), imagers that sense green light (G), and imagers that sense blue light (B). The human eye is more sensitive to green light than to red and blue light; therefore, an increase in the resolution of a synthesized image can be achieved using an array that includes more imagers that sense green light than imagers that sense red or blue light.

[0098] A 5×5 imager array (210) including 17 imagers that sense green light (G), four imagers that sense red light (R), and four imagers that sense blue light (B) is illustrated in FIG. 2B. In several embodiments, the imager array also includes imagers that sense near-IR wavelengths, which can be used to improve the performance of the array camera in low light conditions. A 5×5 imager array including near-IR sensors is illustrated in FIG. 2C. The imager array 220 includes four imagers that sense red light (R), four imagers that sense blue light (B), and two imagers that sense green light. The remaining fifteen imagers indicated with an X can be either imagers that sense green light or imagers that sense near-IR wavelengths. In many embodiments, the distribution of imagers that sense green light and imagers that sense near-IR wavelengths is determined so that distribution is symmetrical.

[0099] In many embodiments of the invention, the imager array includes extended color (EC) imagers. As will be discussed further below, an extended-color channel is a channel that is designed to include both non-visible wavelengths and visible wavelengths. In many embodiments, the extended-color channel extends from green spectral wavelengths to the near-IR. A 5×5 imager array including extended color (EC) sensors is illustrated in FIG. 2D. The imager array 230 includes four imagers that sense red light (R), four imagers that sense blue light (B), nine imagers that sense green light (G), and eight imagers that sense extended-color (EC).

[0100] Although specific configurations for imagers that sense blue light (B) and imagers that sense red light (R) are illustrated, these imagers can be switched with respect to each other. Additional imager array configurations are disclosed in U.S. Provisional Patent Application Ser. No. 61/281,662 to Venkataraman et al., the disclosure of which is incorporated by reference herein in its entirety.

[0101] Each imager in the imager array captures a low resolution (LR) image. Image information captured by imagers, whose sensitivity is tuned by some means (e.g., color filters) to the same portion of the wavelength spectrum, can be considered as constituting a channel of imaging information. When different imagers are tuned to different portions of the wavelength spectrum, then multiple channels of image information are produced that can be provided as inputs to a super-resolution (SR) process in accordance with embodiments of the invention. In several embodiments, the imager array captures a single channel of information (B/W), or 3 channels of information (RGB). In many embodiments, the imager array captures the 3 RGB channels and a fourth near-IR channel that can also be used during SR processing to produce a synthesized higher resolution image.

[0102] Although specific imager array configurations are disclosed above, any of a variety of regular or irregular layouts of imagers including imagers that sense visible light, portions of the visible light spectrum, near-IR light, other portions of the spectrum and/or combinations of different portions of the spectrum can be utilized to capture LR images that provide one or more channels of information for use in cross-channel fusion and SR processes in accordance with embodiments of the invention. The processing of captured LR images is discussed further below.

Image Processing Pipeline for Super Resolution

[0103] U.S. patent application Ser. No. 12/967,807 incorporated above discloses super resolution (SR) processes for generating a high resolution (HR) image from low resolution (LR) images captured by an array camera. The processing of LR images to obtain an HR image in accordance with embodiments of the invention typically occurs in an array camera's image processing pipeline. An image processing pipeline in accordance with embodiments of the invention is illustrated in FIG. 3. In the illustrated image processing pipeline 300, pixel information is read out from imagers 100 and is provided to a photometric conversion module 304 for photometric normalization using the captured LR images and photometric calibration data. The output photometrically normalized images are provided to a parallax detection module 308 and to a super-resolution module 314.

[0104] Prior to performing SR processing, the image processing pipeline detects parallax (or disparity), which becomes more apparent as objects in the scene captured by the imager array approach the imager array, using the parallax detection module 308. Parallax information can be provided to the super-resolution module 314 in the form of parallax corrections (scene dependent geometric displacements) and occlusion maps (showing where content is captured in some cameras but not other cameras due to parallax effects).

[0105] Once the parallax information has been generated, the parallax information and the photometrically normalized LR images are provided to the super-resolution module 314 for use in the synthesis of one or more HR images 316. In many embodiments, the super-resolution module 314 performs scene dependent geometric corrections using the parallax information and scene independent geometric corrections using geometric calibration data 306 obtained via the address conversion module 302.

[0106] The LR images acquired by an imager array in accordance with an embodiment of the invention are typically afflicted by geometric distortions inherent to the optics and manufacturing processes used to fabricate the imager array, which are not scene dependent, and parallax distortions, which are scene dependent. SR processes can use information about initial scene-independent geometric corrections of the LR images, and about correcting the scene-dependent geometric distortions introduced by parallax in the LR images. The scene-independent geometric corrections and the scene-dependent geometric corrections (i.e., parallax) are referred to as the geometric corrections for the LR images. Any of a variety of spatial registration approaches can be used to determine the information about the scene-independent geometric corrections for the LR imagers, and typically this portion of the spatial registration of the image content of the acquired LR images is generated using the output of well-known geometric calibrations techniques, which can be performed “off-line”.

[0107] The photometrically normalized and geometrically registered LR images are then utilized in the synthesis of an HR image. The synthesized HR image 316 may then be fed to a downstream color processing module 564, which can perform color correction and/or chroma level adjustment.

Super Resolution Module

[0108] In several embodiments, the SR process is initialized by performing a pilot fusion process that joins (or places) and processes data from multiple input LR images onto a higher resolution output grid to produce a first fused image. The pilot fusion process can involve filling missing samples on the output grid and filtering potentially misplaced samples. A SR process can use the pilot fused image along with the original LR images to perform an optimal image fusion and restoration according to a predetermined objective function.

[0109] An SR module disclosed in U.S. patent application Ser. No. 12/967,807 is conceptually illustrated in FIG. 4. The SR module 314 receives as inputs photometrically normalized image data, in possibly multiple spectral or color channels, scene-dependent geometric correction information (parallax information) and scene-independent geometric correction information (geometric calibration data). Through a sequence of operations, the SR module generates a higher resolution image corresponding to some or all of the input spectral channels.

[0110] The SR module 314 includes a Pilot Fusion module 402, a Dynamic Focus module 404, a Maximum A Posteriori Super Resolution (MAP SR) module 406, and a Cross Channel Processing module 408. The Pilot Fusion module 402 performs a pilot fusion process using parallax information and either photometrically normalized LR images and geometric calibration data, or photometrically normalized and geometrically registered LR images. The term fusion can be used to refer to a process by which image data samples from multiple images having lower resolution are processed and placed appropriately onto a generally higher resolution grid. This is done such that the LR image samples are placed appropriately onto the output grid (using the geometric corrections—i.e. scene independent geometric calibration data and scene dependent parallax information), and such that positions on the grid where there are missing samples get filled, and a filtering of potentially misplaced samples (outliers) occurs. The geometric correction information provided to the SR module is sufficient to spatially register all, or subsets, of the captured LR images in a common system of coordinates and can use a reference system of coordinates, which can be that of an arbitrarily or purposefully-picked image from the LR image set of a channel.

[0111] The Dynamic Focus module 404 may determine a high resolution depth map, or an equivalent representation of it, using the parallax information of the LR samples transferred onto the fusion grid, along with information about the geometry of the imager (e.g., inter-LR image baselines). The Dynamic Focus module uses the HR depth map to determine the set of pixels on which subsequent blocks in the SR processing pipeline operate (i.e., the pixels containing information concerning the “in-focus” areas of the HR image estimate).

[0112] The MAP SR module 406 uses a maximum a posteriori (MAP) estimate to obtain good estimates for pixels of a synthesized HR image based upon at least the observed LR images and, in many instances, other observed information and/or prior constraint information. The SR process can estimate an HR image for a single channel such as the Green channel based upon at least the fused image output by the Pilot Fusion module 402, the LR image data (raw, photometrically normalized or photometrically normalized and geometrically registered LR images), geometric correction information including an occlusion map, and an imaging degradation model (discussed below) that can include an imaging prior (where available). The HR image estimate for the first channel (e.g. the Green channel) can then be used in cross channel processes to estimate SR images for the other channels and to synthesize a full color SR image.

[0113] The Cross Channel Processing module 408 estimates HR images for the Blue and Red color channels using the HR Green channel estimate generated by the MAP SR module 406. The output of the Cross Channel Fusion module is a synthesized HR image including color information from multiple channels.

Pilot Fusion Processes

[0114] A pilot fusion disclosed in U.S. patent application Ser. No. 12/967,807 is illustrated in FIG. 5. The process 500 includes placing (502) the LR image samples from each input LR image on a fusion grid in accordance with geometric correction information (i.e. scene independent geometric calibration data and scene dependent parallax information forming the total pixel shift) associated with the input LR images.

[0115] Placement of LR image samples on a fusion grid in accordance with an embodiment of the invention is illustrated in FIG. 6. In the illustrated embodiment, the pilot fusion fuses three LR images 604 intra-channel (i.e., using the image data only from the fused channel) and the first step in the pilot fusion process is to place the LR image samples 606 onto the fusion grid 602 using the geometrical correction information provided at its input. The LR image samples 606 from each of the first, second, and third LR images are indicated on the fusion grid 602 as circles, triangles, and squares respectively. The first LR image (i.e. “Frame 1”) is taken as the reference image (geometrically), and the other LR images are transferred onto the fusion grid 602.

[0116] Referring again to FIG. 5, outlier image samples are then identified (504) and removed. Processes for identifying outliers typically involve identifying pixels that possess very low levels of correspondence with neighboring pixels. Once outliers are removed from the fusion grid, the remaining image samples can be interpolated (506) to determine values at all fusion grid positions. In many embodiments, the pilot fusion generates a regularly sampled, initial estimate of the higher resolution image sought for by the SR processing, onto the fusion grid 602. This estimate represents the input to the next SR Processing module 406, which performs the super-resolution restoration processing. Pilot fusion can also be used in a cross-channel fusion process to produce initial fused images in each of a visible imaging channel and a non-visible imaging channel.

Super Resolution Processing

[0117] The MAP SR module 406 uses a statistical estimation approach to determine the super-resolved images for each output spectral channel. An SR process disclosed in U.S. patent application Ser. No. 12/967,807 is illustrated in FIG. 7. The process 700 obtains (702) as its inputs the pilot fused image, the photometrically normalized LR image data for each input spectral channel, geometric correction information, and an imaging prior if available. The pilot fused image is selected (704) as an initial estimate in an optimization process that seeks to obtain (706) optimal HR image estimates given the input LR images captured by the imager array and a forward imaging transformation (built based on an image degradation model). In a number of embodiments, the SR process utilizes a proprietary form of a maximum a posteriori (MAP) objective criterion formulation, which is then optimized with respect to the SR image to be estimated for at least one output spectral channel. Objective criterion formulations utilizing a specific forward imaging transformation that describes image degradation are discussed in greater detail in U.S. patent application Ser. No. 12/967,807 incorporated by reference further above.

[0118] A MAP process can involve iteratively refining an estimate of an HR image using an gradient descent technique to obtain an optimal SR image estimate(s). A process for using gradient descent to obtain optimal SR image estimates in accordance with an embodiment of the invention is illustrated in FIG. 8. The process 800 includes initializing (802) the first estimate with the output of the pilot fusion process described previously. The gradient is determined (804) and a new estimate obtained (806). The process can utilize a gradient ∇_xcJ(x^c) of the form:
∇_xcJ(x^c)=VT(x^c)+ψ_Λ1^intra(x^c)+Φ_Λ2 ^inter(x^c,{circumflex over (x)}^c′),  (1)

[0119] as described in U.S. patent application Ser. No. 12/967,807. The process can either iterate until a predetermined optimality criterion is satisfied (808) or for a finite number of iterations.

SR Processes Incorporating Cross-Channel Fusion

[0120] SR processes in accordance with embodiments of the invention can exploit information from visible and non-visible (typically infra-red) imaging channels to facilitate the generation of a synthesized image, which has higher resolution than any of the individual input channel images. In a number of embodiments, the visible and non-visible channels are distinct. In other embodiments, the visible and non-visible channels possess some degree of overlap. The acquisition of the visible and non-visible portions of the spectrum by an imaging device produces data that has different characteristics (such as blur levels, imaging noise, or absolute and relative amplitude responses in the image). In addition, each channel can have a native image resolution that is not necessarily the same between channels. In many embodiments that utilize visible and non-visible channels in SR processing, the visible and non-visible channels are treated jointly and processes are performed that normalize and transfer the LR image samples from each channel onto a common (higher) resolution grid.

[0121] A process for performing cross-channel normalization and fusion of two input-channels covering different portions of the wavelength spectrum (i.e. visible and non-visible) in accordance with an embodiment of the invention is conceptually illustrated in FIG. 9. The process receives as inputs multiple LR images for each channel, where each LR image is acquired at a given resolution, and spatial registration information. In many embodiments, the process also receives as additional input information concerning characteristics of the imaging conditions including, but not limited to, information indicative of light conditions such as the imager analog gain and noise information. The process performs (902) a pilot fusion process on each channel that is similar to the pilot fusion process described above. During the pilot fusion process, the LR image samples of a channel are fused onto an intermediate grid. Each channel's intermediate grid has the same resolution, which is typically higher than that of the input LR images. As discussed above, pilot fusion processes in accordance with embodiments of the invention can utilize interpolation and filtering processes that are locally adaptive and, which in many embodiments, use information about the noise statistics of the acquiring sensor, as well as imaging condition information including but not limited to the analog gain of the imagers.

[0122] The process 900 includes spatial registration (904) between the two fused channel images. Given differences between the information in each of the two channels, many embodiments of the invention determine this registration using the higher resolution intermediate grids. Attempting to spatially register the LR images of the two channels in their native (lower) resolutions can be more difficult. Any of a variety of known spatial registration processes appropriate to the application can be utilized. In general, however, the differences between the information contained in visible and non-visible channels means that feature-based spatial registration processes tend to provide the best spatial alignment between the non-visible channel fused image and the visible channel fused image.

[0123] As a pre-cursor to cross-channel normalization, the fused images are denoised (906). In a number of embodiments, filtering operations are applied cross-channel (i.e., between the visible and non-visible spatially registered channels). In several embodiments, bilateral filters are used to perform local filtering operations. As noted above, the weights of bilateral filters are a function of both the photometric and geometric distance between a pixel and the pixels in its neighborhood. In a number of embodiments, a cross-channel bilateral filter is first applied to the visible channel fused image, using weights (or the local kernels) determined for the bilateral filter that would normally be applied to the non-visible fused image. The bilateral filter is applied in this way to exploit the advantageous properties of the non-visible channel. For example, when the non-visible channel is a near-IR channel, the channel generally has a higher SNR in low light conditions. In many embodiments, the cross-channel filtering is applied conditionally based upon imaging conditions. Thus, for low light conditions (e.g., defined by the value of the imager analog gain), a cross-channel bilateral filter is applied. For higher light conditions, however, a conventional bilateral filter whose weights are determined based on the visible-channel image (i.e. no longer using information from the other channel) is applied. In a number of embodiments, both the cross-channel and conventional bilateral filtering processes are done locally and are adapted using available noise statistics information. The non-visible channel fused image is also denoised (906). In a number of embodiments, denoising is performed using a conventional bilateral filter (i.e. using weights determined based on the non-visible channel fused image). In many embodiments, the bilateral filtering of the non-visible channel denoised image is also adaptive using available noise statistics information.

[0124] A process for applying bilateral filters to denoise images in accordance with embodiments of the invention is illustrated in FIG. 10A. The process includes determining (1002) if there is a very low light condition. If there is very low light, a bilateral filter is applied (1004) to the non-visible channel. The process also includes determining (1006) if there is a low light condition. If there is low light, a cross-channel bilateral filter is applied to the visible channel using weights from the non-visible channel. Although specific cross-channel denoising filters are disclosed above, any of a variety of filtering and cross-channel filtering techniques can be used to denoise the non-visible and visible fused images as appropriate to a specific application in accordance with embodiments of the invention.

[0125] Referring again to FIG. 9, one fused image is cross-channel normalized to the other fused image (908). In a number of embodiments, the normalization is performed by generating gain and offset values for pixels in the image to be normalized.

[0126] In several embodiments, the denoised non-visible channel fused image is normalized in the photometric reference space of the denoised visible channel fused image by applying gains and offsets to pixels of the denoised non-visible channel fused image. For a neighborhood window around an EC pixel and the corresponding G pixel, determine the gain and bias required to adjust the EC pixel to the local characteristics of the G channel. The gain and offset for each normalized EC pixel can be determined using the following equations:

[0127] $\begin{array}{cc}\stackrel{\_}{e}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}e\left(r,c\right),& \left(2\right)\\ \stackrel{\_}{g}=\frac{1}{N_r{N}_c}\sum _r^{}\sum _c^{}g\left(r,c\right),& \left(3\right)\\ \hat{a}=\frac{\lfloor \sum _r^{}\sum _c^{}g\left(r,c\right)·e\left(r,c\right)\rfloor -N_r{N}_c\stackrel{\_}{g}·\stackrel{\_}{e}}{\left[\sum _r^{}\sum _c^{}{e}^2\left(r,c\right)\right]-N_r{N}_c{\stackrel{\_}{e}}^2},& \left(4\right)\\ \hat{b}=\stackrel{\_}{g}-\hat{a}·\stackrel{\_}{e},& \left(5\right)\\ e_n\left(i,j\right)=\hat{a}·e\left(i,j\right)+\hat{b}& \left(6\right)\end{array}$

[0128] where e=the image to be normalized (EC channel), g=the reference image (green channel), N_r and N_c are the number of pixels horizontally and vertically of the neighborhood of pixels around the pixel being computed (EC), and r and c are row and column indices into the images within the bounds defined by N_r and N_c. The computed {circumflex over (α)} is the gain and {circumflex over (b)} is the offset for each pixel e_n(i, j) to be normalized. In further embodiments, a joint-bilateral-style normalization can be utilized determine a subset of pixels to normalize using the above equations as will be discussed further below. In other embodiments, any of a variety of normalization techniques can be applied as appropriate to a specific application. Moreover, in different applications (such as low light “night vision”) the visible channel fused image can be normalized to the non-visible channel fused image.

[0129] In many embodiments, the denoised fused images are discarded following the generation of the gains and offsets and the original non-visible fused image (i.e., the image prior to denoising) is normalized photometrically to the reference space of the original fused visible channel image (i.e., the image prior to denoising).

[0130] Once the two fused images are normalized, the two normalized fused images are fused (910) onto a common higher resolution grid using a pilot fusion process similar to the process used to initially create the two fused images.

[0131] In the process described above, if the output fused image is intended for human viewing, information from the non-visible channel may not be desired in the output image. Therefore, the process can be disrupted to ignore predetermined features or information from the non-visible image. For example, features that exist in the non-visible fused image, and do not exist in the visible image, or vice-versa, can disrupt the photometrical normalization of the images. In a number of embodiments, local features are compared in the two fused and spatially-registered channel images using a bilateral filter weight determination, to: a) eliminate the undesired information existing in the non-visible channel, and prevent it from being transferred into the visible channel; and b) to prevent the improper determination of cross-channel bilateral filter weights to be applied in the visible channel, due to features existing in the non-visible channel and missing in the visible channel. In other embodiments, a joint bilateral filter is used that does not transfer edges from one source channel to the output. Any of a variety of techniques can be utilized to exclude non-desired information from the non-visible channel appropriate to a specific application in accordance with embodiments of the invention. A process for joint-bilateral-style-normalization that can be utilized to determine a subset of pixels to be cross-channel normalized is discussed below.

Joint-Bilateral-Style Normalization of Diverse Spectral Channels with Edge Preservation

[0132] The response of different spectral channels to a particular reflected color can be similar or completely uncorrelated. Adjoining different colored areas might exhibit an edge or might be perceived as the same and exhibit no hint of an edge. The one constant is that the relative response to a given color remains the same throughout a scene.

[0133] In order to utilize two or more spectral channels for purposes of SR, the levels of response to a given color can be adjusted to be the same. Generally adjusting the levels is simply and effectively accomplished using a neighborhood-based gain and offset normalization, as described above. However, in areas of strong edges due to highly contrasting colors (i.e. highly contrasting response to particular colors) information from areas that do not contain the pixel of interest can cause artifacts along edges to a distance from the edge equal to half the window width. In order to constrain the information in the filter to areas consistent with the pixel of interest a bilateral-filter approach can be employed.

[0134] Bilateral filters have excellent edge-preserving capabilities as pixels similar in quality (level/color, distance, etc.) to the pixel of interest are the most highly weighted in the filter calculations. Gain and offset normalization, however, does not apply a convolution-style filter but rather uses data in a neighborhood of the pixel-of-interest and the neighborhood of its corresponding pixel in the training channel to find a best (least-squares) estimate of a bias and gain adjustment for the pixel of interest. A joint bilateral filter style approach can be employed to restrict the estimation to be based on only those pixels that meet a similarity requirement. The proximity of the POI to an edge or its corresponding pixel to an edge is complicated by the case of a given edge appearing in one channel but not in the other. An additional complication is that an edge that appears in a non-visible-light spectral channel but not in a visible light channel should not be suppressed for normal imaging (but perhaps for certain hyper-spectral imaging cases).

[0135] A process for joint-bilateral-style normalization of one spectral channel to another spectral channel in accordance with embodiments of the invention is illustrated in FIG. 10B. In several embodiments of the invention, the joint bilateral filter style normalization of two channels involves selecting (1052) a pixel of interest in the first channel and a collection of similar pixels in the neighborhood of the pixel of interest. Without loss of generality, the first channel can be assumed to be registered with less than ½ pixel difference to the second channel. Similarity of pixels can be defined in a bilateral-filter-style manner where the resulting value is thresholded to make a binary decision whether to include or exclude a particular pixel from the neighborhood calculation. In many embodiments, similarity criteria also considers signal to noise ratio.

[0136] In the second channel, select (1054) a pixel of interest corresponding to the first pixel of interest and a collection of pixels similar to that corresponding pixel of interest. Because colors in a scene are often mapped to the same level or value in a spectral channel, it is sometimes the case that the two collections of pixels do not have a positional one-to-one correspondence. However, one collection should usually be a subset of the other (modulo noise and difference due to subpixel-level misregistration). The intersection of the two collections of pixels (i.e., pixels in common) is then determined (1056) and used to calculate (1058) the gain and offset estimates using equations (2) to (6) above. In many embodiments, the denoised non-visible channel fused image is the first channel normalized to the denoised visible channel fused image which is the second channel. However, any image can be used as the first and second channel as appropriate to a specific application. For example, in an application using “night vision” in ultra low light conditions, a visible channel (e.g., green) may be normalized to a non-visible channel. In some applications, an “extended-color” channel includes visible and non-visible wavelengths, and can offer advantages over a strictly non-visible channel when used in place of a non-visible channel in some of the image processing methods discussed above. The use of “extended-color” channels is discussed below.

Use of “Extended-Color” Channels

[0137] Use of non-visible channels of information during image fusion and super-resolution in accordance with embodiments of the invention can exploit certain characteristics of the non-visible channels, such as sensitivity in low light imaging conditions. Such requirements can be satisfied by a near infra-red channel that can supplement the information captured in visible-light channels. However, such diversity of the channels in the context of super-resolution can introduce additional difficulties in fusing the information together to achieve the desired super-resolution factor increase. The difficulties are primarily related to the challenges associated with spatially registering with fine precision the spectrally-diverse channel images, and also the problem of normalizing the photometric responses (e.g., absolute and relative signal magnitudes) across the two channels, such that the overall information can be properly fused onto a common grid. In many embodiments, a cross-channel fusion is performed using a so-called “extended-color” channel in place of the non-visible channel. An extended-color channel is a channel that is designed to overlap spectrally with the visible channel spectrum. In many embodiments, the extended-color channel extends from green spectral wavelengths to the near-IR. The channel response of an extended-color channel in accordance with embodiments of the invention is illustrated in FIG. 11A. As can be seen from FIG. 11A, the channel response 1102 of the extended-color channel encompasses green 1104, red 1106, and near-IR spectral wavelengths.

[0138] The channel response of an alternative extended-color channel in accordance with embodiments of the invention is illustrated in FIG. 11B. As can be seen from FIG. 11B, the channel response 1152 of the extended-color channel encompasses green 1154, red 1156, and near-IR spectral wavelengths. Although specific channel responses are illustrated in FIGS. 11A and 11B, extended-color channels can have any of a variety of channel responses encompassing both visible and non-visible spectral wavelengths in accordance with embodiments of the invention.

[0139] An extended-color channel can be used in place of a non-visible channel in the processes discussed above and can be used in a cross-channel fusion process with a visible channel as discussed below.

[0140] Cross-channel fusion of a visible channel, such as a green (G) channel, and an extended-color channel in accordance with embodiments of the invention can provide significant benefits during SR processing. Specifically, the desirable characteristics of a near-IR channel are retained (i.e., sensitivity in low light imaging situation), however, the spectral overlap between the extended-color channel and the visible channel significantly increases the correlation between the two channels. The increased correlation can make spatial registration and photometric normalization between the visible channel and the extended-color channel much more robust. For example the magnitude, location, and even existence of edges in the images of the two channels will be much more correlated, and easier to exploit in the SR processing. The overlay with the visible spectrum in the extended color channel definition also provides a degree of safety for the SR processing in cases where the response of the pure near-IR channel (i.e. not overlapping with the visible spectrum) would be very weak due (e.g., due to the characteristics of the materials in the scene) thus reducing the SR factor that could be achieved, by reducing the number of images available for SR processing.

[0141] For the reasons outlined, array cameras in accordance with many embodiments of the invention include imager arrays incorporating extended-color imagers. A variety of imager array layouts can be utilized including a layout similar to the layout shown in FIG. 2D. The 5×5 imager array 230 includes four imagers that sense red light (R), four imagers that sense blue light (B), nine imagers that sense green light (G), and eight imagers that sense extended-color (EC). In other embodiments, extended-color imagers can be used in one dimensional array cameras, in different two dimensional array camera configurations to the configuration shown in FIG. 2D, and/or in arrays of array cameras.

Cross-Channel Fusion Processing Pipeline

[0142] Processes for performing cross-channel fusion are discussed above between visible and non-visible channels and between a visible channel and an extended-color channel. A cross-channel fusion processing pipeline in accordance with an embodiment of the invention is illustrated in FIG. 12. The SR processing module 1200 accepts as inputs LR images from a first channel, which is usually a visible channel, LR images from a second channel, which is typically a non-visible or extended color channel, registration information, and information concerning the imaging conditions such as, but not limited to, noise information and the analog gain of the imagers used to capture the LR images. The LR images from the first channel and the associated registration information is provided to a first pilot fusion module 1202 and the LR images from the second channel and the associated registration information is provided to a second pilot fusion module 1204. Both pilot fusion modules perform a pilot fusion process similar to the pilot fusion processes described above which fuse the image samples onto an intermediate grid. The intermediate grid is typically higher resolution than the native resolution of the LR images. In a number of embodiments, the resolution of the intermediate grid is L times the height and width of the LR images. The fused images for each channel are provided to a spatial registration module 1206, which spatially registers the two fused images prior to the denoising of the two images. In many embodiments, a first filtering module 1208 performs either cross-channel or bilateral filtering on the first channel fused image depending upon the light conditions when the LR images were captured. A second filtering module 1210 applies a bilaterial filter to the second channel fused image. The denoised images are provided to a cross-channel normalization module 1212, which normalizes the images with the aid of locally adaptive logic 1214. Locally adaptive logic can include criteria for the adjustment of filter windows, strengths, and/or thresholds based on characteristics of the local signal such as local SNR signature. For example, a larger filter window can be used at lower SNR.

[0143] As noted above, the cross-channel normalization module can photometrically normalize the denoised images or can photometrically normalize the original fused images. The output of the cross-channel normalization module is provided to a cross-channel fusion module 1216, which fuses the fused image from each channel onto an output grid using a pilot fusion process similar to the pilot fusion processes described above.

[0144] Although a specific cross-channel fusion processing pipeline is illustrated in FIG. 12, alternative processing pipelines that fuse information from different channels in a manner acceptable to a specific application can be utilized in accordance with embodiments of the invention.

[0145] While the above description contains many specific embodiments of the invention, these should not be construed as limitations on the scope of the invention, but rather as an example of one embodiment thereof. It is therefore to be understood that the present invention may be practiced otherwise than specifically described, without departing from the scope and spirit of the present invention. Thus, embodiments of the present invention should be considered in all respects as illustrative and not restrictive.