Document segmentation for mixed raster content representation

[0001] 1. Field of the Invention

[0002] The present application relates to compression techniques and, more particularly, improved compression techniques for use with the mixed raster content compression format.

[0003] 2. Description of the Related Art

[0004] Image compression addresses the problem of reducing the amount of data required to represent a digital image. Compressed data (i.e., text and/or image data) may be easier to transmit by, for example, facsimile and/or e-mail than non-compressed data.

[0005] The underlying basis of the reduction process is the removal of redundant data during a scanning operation or, alternatively, from existing electronic files. Compression techniques typically fall into two broad categories: information preserving compression (i.e., lossless compression) and lossy compression. Lossless compression allows data to be compressed and decompressed without the loss of information. Lossy compression provides higher levels of data reduction as compared to lossless compression but results in a less than perfect reproduction of the original image (i.e., information loss).

[0006] Lossless compression methods (e.g., Lempel-Ziv or LZW) perform well on text but do not perform particularly well on scanned images. In contrast, lossy compression methods (e.g., JPEG) work fairly well on continuous-tone pixel arrays but do not work particularly well on text.

[0007] Accordingly, the mixed raster content (MRC) format may be used to compress documents including both text and images. The MRC format segments a document into two or more planes or layers. For example, each pixel in a document may be segmented into a text layer, a foreground layer and a background layer, as described in U.S. Pat. Nos. 5,778,092 and 6,324,305, the entire contents of which are incorporated herein by reference. The text layer may include only text data of the document, the foreground layer may include color information of the text data and the background layer may include any image that is not considered text data.

[0008] Once each pixel has been segmented between two or more layers, each layer may be separately compressed using an appropriate compression method based on the content of the layer. For example, the text layer may be compressed using a lossless compression method such as LZW while the foreground and background layers may be compressed using a lossy compression method such as Joint Photographic Experts Group (JPEG).

[0009] In the prior art, the segmentation step may be carried out by separating high frequency pixels, which typically are associated with text, from low frequency pixels, which typically are associated with image and background data. Such edge detection techniques often misclassify pixels belonging to large halftone dots (i.e., halftone dots having a size equal to or greater than the edge detector kernel) as high frequency pixels and therefore treat them as text pixels, thereby increasing the size of the compressed file.

[0010] Accordingly, there is a need for a segmentation technique having an improved capability for detecting the difference between halftone areas and text areas.

[0011] In one aspect, the segmentation method provides a method for segmenting a selected pixel of an array of pixels between at least two layers. The method includes identifying an N-by-N window centered upon the selected pixel, evaluating at least one pixel in the N-by-N window to determine whether the selected pixel is a potential text element, identifying an M-by-M window centered upon the selected pixel when the act of evaluating determines that the selected pixel is a potential text element, wherein the M-by-M window is smaller than the N-by-N window, and determining whether the potential text element includes text by comparing at least two pixels within the M-by-M window.

[0012] In another aspect, the segmentation method provides a method for segmenting a selected pixel of an array of pixels between at least two layers. The method includes identifying a N-by-N window centered on the selected pixel, obtaining a total intensity value for each row, column and major diagonal of the N-by-N window, comparing the total intensity values to determine whether the selected pixel is a potential text element, identifying a M-by-M window centered on the selected pixel when the selected pixel is determined to be a potential text element, wherein the M-by-M window is smaller than the N-by-N window, and comparing at least two oppositely positioned pixels within the M-by-M window to determine whether the potential text element includes text.

[0013] Other aspects of the segmentation method will become apparent from the following description, the accompanying drawings and the appended claims.

[0016] As shown in FIG. 1, a document may be represented as a pixel array 100 having X number of rows and Y number of columns. For example, as shown in FIG. 1, the pixel array 100 may include 19 rows of pixels (i.e., rows 1 to 19 in FIG. 1) and 19 columns of pixels (i.e., columns A to S in FIG. 1) for a total of 361 pixels. Thus, when a document is being segmented and compressed, each pixel must be segmented to one of at least two layers. In one aspect, each pixel may be segmented to either a background layer, a foreground layer or a text layer. In another aspect, each pixel may be segmented to either a background layer or a text layer.

[0017] As shown in FIG. 2, the segmentation method, generally designated 110, may begin at box 112. As shown at box 114, a pixel may be selected from the pixel array 100 for evaluation and segmentation. For example, referring to FIG. 1, the segmentation method 110 may select pixel J-10 (i.e., the pixel located at row 10 of column J), designated as pixel 130, from pixel array 100 for evaluation and segmentation. Once pixel 130 has been selected, the segmentation method 110 may determine whether the selected pixel 130 is a potential text element (PTE), as shown in boxes 116 and 118 of FIG. 2. A PTE may be a pixel that potentially includes text or a pixel that may ultimately be classified as text. If the method 110 determines that the selected pixel 130 is not a PTE, the pixel 130 may be segmented to the background layer, as shown in box 120, and the method 110 may select another pixel for evaluation and segmentation, as shown in box 114. If the method 110 determines that the selected pixel 130 is a PTE, the method 110 may further evaluate the pixel, as shown in boxes 122 and 124, to determine whether the pixel 130 includes text. If the method 110 determines that the selected pixel 130 includes text, the pixel 130 may be segmented to the text layer, as shown in box 126, and the method 110 may proceed to the next pixel, as shown in box 114. If the method 110 determines that the pixel 130 does not include text, the pixel 130 may be segmented to the background layer, as shown in box 120, and the method 110 may proceed to the next pixel, as shown in box 114. Thus, the method 110 may evaluate and segment each pixel in the array 100.

[0018] The first act in the process (i.e., PTE detection), as shown in boxes 116 and 118, may include determining whether the selected pixel is a PTE. PTE detection may begin by identifying an N-by-N window 132 around the selected pixel 130. “N” may be any number and may be selected by the user or may be pre-programmed. “N” should be sufficiently large to allow the method 110 to identify halftone dots, edges and the like. In one aspect, “N” may be a number between about 5 and about 500.

[0019] For example, as shown in FIG. 1, the selected pixel 130 may be pixel J-10 and N may equal 17. Thus, the window 132 (i.e., a 17 by 17 window) may include 17 rows (i.e., rows 2 to 18) and 17 columns (i.e., columns B to R) centered on pixel 130 (i.e., pixel J-10).

[0020] Each pixel in the array 100 may have an associated intensity value (i.e., a pixel intensity), as shown by numbers 50 and 250 throughout FIG. 1. For example, the intensity value may be a number between about 0 and about 255. In one aspect, the method 110 may directly use each pixel's color value (e.g., red, green, blue or grey color values). In another aspect, the method 110 may derive a pixel's intensity value from its color values.

[0021] Once a window 132 around selected pixel 130 has been identified, the method 110 may proceed to sum the intensity values of each row of pixels (i.e., a summation of the pixel intensity of each pixel in a row for each row in the window 132), each column of pixels (i.e., a summation of the pixel intensity of each pixel in a column for each column in the window 132) and each major diagonal of pixels (i.e., a summation of the pixel intensity of each pixel in a diagonal for each major diagonal in the window 132).

[0022] For example, as shown in FIG. 1, the summation of the intensity values for each row of window 132 (i.e., rows 2 to 18) are provided in Table 1:

[0023]

	TABLE 1
	Row	Total Intensity
	2	850
	3	850
	4	850
	5	850
	6	850
	7	850
	8	850
	9	850
	10	850
	11	850
	12	850
	13	3250
	14	1050
	15	1050
	16	1050
	17	1050
	18	1050

The summation of the intensity values for each column of window 132 (i.e., columns B to R) are provided in Table 2:

[0024]

	TABLE 2
	Column	Total Intensity
	B	1050
	C	1050
	D	2050
	E	1050
	F	1050
	G	1050
	H	1050
	I	1050
	J	1050
	K	1050
	L	1050
	M	1050
	N	850
	O	850
	P	850
	Q	850
	R	850

The summation of the intensity values for each major diagonal of window 132 (i.e., diagonal D₁ is pixel B-2 to pixel R-18 and diagonal D₂ is pixel R-2 to pixel B-18) are provides in Table 3:

[0025]

	TABLE 3
	Diagonal	Total Intensity
	D1	1050
	D2	1250

[0026] Once the total intensities for each row and column have been obtained, the method 110 identifies a maximum (i.e., M_r) and a minimum (i.e., m_r) total intensity of all the rows and a maximum (i.e., M_c) and a minimum (i.e., m_c) total intensity of all the columns and makes a comparison, as follows:
M_r−m_r≦T₁  (1)
M_c−m_c≦T₁  (2)
wherein T₁ is a predetermined threshold value. Furthermore, the method compares the total intensity values of the major diagonals, as follows:
|D₁−D₂|≦T₂  (3)
wherein T₂ is a predetermined threshold value. The threshold values T₁ and T₂ may be selected by the user or may be preprogrammed.

[0027] If each comparison is less than the associated threshold value (i.e., each of equations 1, 2 and 3 are true), the pixel is not a PTE. However, if at least one comparison exceeds the associated threshold value (i.e., at least one of equations (1), (2) and (3) is false), the pixel is considered a PTE.

[0028] For example, using
T₁=1000(predetermined value)  (4)
T₂=500(predetermined value)  (5)
and applying the results of Tables 1, 2 and 3 to equations (1), (2) and (3), the following is obtained:
M_r=3250(per row 13,see Table 1)  (6)
m_r=850(per rows 2-12,see Table 1)  (7)
M_c=2050(per column D,see Table 2)  (8)
m_c=850(per columns N-R,see Table 2)  (9)
D₁=1050(see Table 3)  (10)
D₂=1250(see Table 3)  (11)
Substituting the values for M_r and m_r for (6) and (7) into equation (1) provides:
3250−850≦1000(false)  (12)
Substituting the values for M_c and m_c for (8) and (9) into equation (2) provides:
2050−850≦1000(false)  (13)
Substituting the values for D₁ and D₂ for (10) and (11) into equation (3) provides:
|1050−1250|≦500(true)  (14)

[0029] Thus, equation (1) (i.e., 2400≦1000) and equation (2) (i.e., 1200≦1000) would and equation (3) (i.e., 200≦500) would be true for the results of Tables 1, 2 and 3. Therefore, the pixel 130 (i.e., pixel J-10) may be classified as a PTE, and the method 110 may proceed as shown in boxes 122 and 124.

[0030] The second act (i.e., text detection), as shown in boxes 122 and 124 of FIG. 2, may determine whether the PTE includes text using a modified line detector kernel. If the PTE includes text, the selected pixel may be segmented to the text layer, as shown in box 126. In contrast, if the PTE is not identified as including text, the selected pixel may be segmented to the background layer, as shown in box 120.

[0031] Text detection of the method 110 may begin by identifying an M-by-M window 134 around the selected pixel 130, wherein the M-by-M window 134 is smaller than the N-by-N window 132. “M” may be selected by the user or may be pre-programmed. In one aspect, “M” may be a number between about 2 and about 300.

[0032] For example, as shown in FIG. 1, M may equal 7. Thus, the M-by-M window 134 (i.e., a 7 by 7 window) may include 7 rows (i.e., rows 7 to 13) and 7 columns (i.e., columns G to M) centered on pixel 130 (i.e., pixel J-10).

[0033] The method 110 may proceed to compare intensity values of each pixel in a first row and/or column to a corresponding pixel in an opposite row and/or column to determine whether each pixel in the first row and/or column has an intensity greater than the intensity of the corresponding pixel in the opposite row and/or column by a predetermined threshold value T3.

[0034] In one aspect, the method 110 may only compare the pixels directly bordering the M-by-M window 134. For example, referring to FIG. 1, the method 110 may only compare each pixel in row 7 (i.e., pixels G-7, H-7, I-7, J-7, K-7, L-7 and M-7) to each corresponding pixel in row 13 (i.e., pixels G-13, H-13, I-13, J-13, K-13, L-13 and M-13) and/or each pixel in column G (i.e., pixels G-7, G-8, G-9, G-10, G-11, G-12 and G-13) to each corresponding pixel in column M (i.e., pixels M-7, M-8, M-9, M-10, M-11, M-12 and M-13). In another aspect, the method 110 may compare additional pixels within the M-by-M window 134 to corresponding opposite pixels. For example, each pixel in row 8 (i.e., pixels G-8, H-8, I-8, J-8, K-8, L-8 and M-8) may be compared to a corresponding pixel in row 12 (i.e., G-12, H-12, I-12, J-12, K-12, L-12 and M-12).

[0035] The comparison may ask, for example, whether any of the following is true:
|I_G-7−I_G-13|≧T₃  (15)
|I_H-7−I_H-13|≧T₃  (16)
|II-7−II-13|≧T3  (17)
|IJ-7−IJ-13|≧T3  (18)
|IK-7−IK-13|≧T3  (19)
|IL-7−IL-13|≧T3  (20)
or
|IM-7−IM-13|≧T3  (21)
wherein I_Y-X is the intensity of the pixel located at column Y and row X. If any of equations 4-10 are true, then the selected pixel 130 may be designated as a true text pixel and segmented to the text layer, as shown in box 126 of FIG. 2. However, if none of the pixels in a row (or column) have an intensity greater than the intensities of the corresponding pixels in an opposite row (or column) by a predetermined threshold value T₃ (i.e., each of equations 15-21 are false), then the selected pixel 130 is not considered a text pixel and therefore is segmented to the background layer, as shown in box 120 of FIG. 2.

[0036] Applying equations 15-21 to the selected pixel 130 of FIG. 1 (i.e., pixel J-10) using the pixel intensities listed in FIG. 1 and a threshold value T₃ of 100, the following result for each of equations 15-21 is obtained:
|50−250|≧100  (22)
or
200≧100(true)  (23)
Therefore, pixel J-10 may be designated as a true text pixel and segmented to the text layer.

[0037] If we were to compare each pixel in column G to its corresponding pixel in column M we would obtain a different result. In particular, we would obtain 0≧100 for each comparison (i.e., |50−50| for rows 7 through 12 and |250−250| for row 13), which is false. However, so long as one comparison (either column or row) yields a result that is greater than the threshold T₃, the selected pixel 130 may be segmented as a text pixel.

[0038] Once each pixel in the array 100 is segmented into at least two layers (e.g., a text layer and background layer, as described above), the document may be separated into the at least two layers, and each layer may be compressed using a compression technique appropriate for that layer (e.g., LZW for the text layer and JPEG for the background layer).

[0039] Although the segmentation method is shown and described with respect to certain aspects, it is obvious that modifications will occur to those skilled in the art upon reading and understanding this specification. The segmentation method includes all such modifications and is limited only by the scope of the claims.