DOCUMENT IMAGE COMPRESSION METHOD AND ITS APPLICATION IN DOCUMENT AUTHENTICATION

• US 20140185933A1
• Filed: 12/28/2012
• Published: 07/03/2014
• Est. Priority Date: 12/28/2012
• Status: Active Grant

• Current Assignee: Konica Minolta Laboratory U.S.A., Inc. (Konica Minolta Inc.)
• Original Assignee: Yibin Tian
• Inventors: Tian, Yibin, Ming, Wei
• Granted Patent: US 9,230,383 B2
• Time in Patent Office: Days
• US Class Current: 382/173
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06T 2207/10004 Still image; Photographic image
• CPC Class Codes: G06T 2207/30176 Document
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06V 20/80 Recognising image objects characterised by unique random patterns
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/15 Cutting or merging image elements, e.g. region growing, watershed or clustering-based techniques
• CPC Class Codes: G06V 30/1914 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries, e.g. user dictionaries
• CPC Class Codes: G06V 30/224 of printed characters having additional code marks or containing code marks
• CPC Class Codes: G06V 30/413 Classification of content, e.g. text, photographs or tables
• CPC Class Codes: G07D 7/004 using digital security elements, e.g. information coded on a magnetic thread or strip
• CPC Class Codes: G07D 7/0043 using barcodes
• CPC Class Codes: G07D 7/2033 Matching unique patterns, i.e. patterns that are unique to each individual paper
• CPC Class Codes: G07D 7/2041 Matching statistical distributions, e.g. of particle sizes orientations
• CPC Class Codes: G07D 7/206 Matching template patterns
• CPC Class Codes: G07D 7/2083 Learning
• CPC Class Codes: H04N 1/40062 Discrimination between different image types, e.g. two-tone, continuous tone

Abstract

A method for compressing a bi-level document image containing text is disclosed. The document image is segmented into symbol images each representing a letter, numeral, etc. in the document. The symbol images are classified into a plurality of classes, each class being associated with a template image and a class index. Classification is done by comparing each symbol to be classified with template of existing classes, using a number of image features including zoning profiles, side profiles, topology statistics, and low-order image moments. These image features are compared using a tolerance based method to determine whether the symbol matches the template. After classification, certain classes that have few symbols classified into them may be merged with other classes. In addition, the template images of the classes are down-sampled, where the final sizes of the template images are dependent on the likelihood of confusion of the template with other templates.

Claims

1. A method for compressing a binary image representing a document containing text regions, the method comprising:

(a) segmenting the text regions into a plurality of symbol images, each symbol image representing a symbol of text, each symbol image being bound by a bounding box having a location and a size;

(b) classifying each symbol image obtained in step (a) into one of a plurality of classes, each class being represented by a template image and a class index, including, for each symbol image being classified;

(b1) comparing the symbol image with each template image to determine whether they match each other, including comparing a plurality of features of the symbol image with the corresponding plurality of features of the template image, the plurality of features including density statistics features, side profile features, topology statistics features and shape features;

(b2) if a match is found in step (b1), recording the class index corresponding to the matched template in association with the symbol image being classified; and

(b3) if no match is found in step (b1), adding a new class to the plurality of classes, by using the image of the symbol image being classified as the template image of the new class and assigning a class index to the new class, and recording the class index in association with the symbol image being classified;

(c) resizing the template image of each class to a final size; and

(d) storing, as compressed image data, the resized template image for each of the plurality of classes along with its class index, the bounding box location and size for each symbol image obtained in step (a), and the class index for each symbol image obtained in step (b2) or (b3).

2. The method of claim 1, wherein step (a) includes segmenting the text regions into a plurality of lines, segmenting the lines into a plurality of words, and segmenting the words into the plurality of symbol images.

3. The method of claim 1, wherein the density statistics features include zoning profile features, wherein the shape features include low-order moment features, and

wherein step (b1) includes, for each template image being compared:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the symbol image that are different from corresponding features of the template image; and

wherein the symbol image and the template image are determined to match each other if;

the first difference number is smaller than or equal to a first threshold value, the second difference number is smaller than or equal to a second threshold value, the third difference number is smaller than or equal to a third threshold value, and the fourth difference number is smaller than or equal to a fourth threshold value.

4. The method of claim 1, wherein the density statistics features include zoning profile features, wherein the shape features include low-order moment features, and

wherein step (b1) includes, for each template image being compared:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the symbol image that are different from corresponding features of the template image; and

wherein the symbol image and the template image are determined to match each other if a sum of the first through fourth difference numbers is smaller than or equal to a threshold value.

5. The method of claim 1, further comprising, after step (b) and before step (c):

merging a first one of the plurality of classes into a second one of the plurality of classes if a total number of symbol images classified into the first class in step (b) is below a threshold and a measure of confidence of classification for the first class is below another threshold.

6. The method of claim 1, wherein in step (c), the final sizes for at least some template images are different from each other, wherein step (c) includes:

(c1) calculating a similarity measure between each template image with each other template image;

(c2) determining a final size for each template image based on the calculated similarity measure with other template images; and

(c3) resizing each template image to the final size determined in step (c2).

7. The method of claim 6, wherein step (c1) includes, for each pair of template images:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the pair of template image that are different from each other; and

summing the first through fourth difference numbers as the similarity measure.

8. A method for compressing a binary image representing a document containing text regions, the method comprising:

(a) segmenting the text regions into a plurality of symbol images, each symbol image representing a symbol of text, each symbol image being bound by a bounding box having a location and a size;

(b) classifying each symbol image obtained in step (a) into one of a plurality of classes, each class being represented by a template image and a class index, including, for each symbol image being classified;

(b1) comparing the symbol image with each template image to determine whether they match each other;

(b2) if a match is found in step (b1), recording the class index corresponding to the matched template in association with the symbol image being classified; and

(b3) if no match is found in step (b1), adding a new class to the plurality of classes, by using the image of the symbol image being classified as the template image of the new class and assigning a class index to the new class, and recording the class index in association with the symbol image being classified;

(c) resizing the template image of each class to a final size, wherein the final sizes for at least some template images are different from each other, including;

(c1) calculating a similarity measure between each template image with each other template image;

(c2) determining a final size for each template image based on the calculated similarity measure with other template images; and

(c3) resizing each template image to the final size determined in step (c2); and

(d) storing, as compressed image data, the resized template image for each of the plurality of classes along with its class index, the bounding box location and size for each symbol image obtained in step (a), and the class index for each symbol image obtained in step (b2) or (b3).

9. The method of claim 8, wherein step (c1) includes, for each pair of template images:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the pair of template image that are different from each other; and

summing the first through fourth difference numbers as the similarity measure.

10. The method of claim 8, further comprising, after step (b) and before step (c):

merging a first one of the plurality of classes into a second one of the plurality of classes if a total number of symbol images classified into the first class in step (b) is below a threshold and a measure of confidence of classification for the first class is below another threshold.

11. A computer program product comprising a computer usable non-transitory medium having a computer readable program code embedded therein for controlling a data processing apparatus, the computer readable program code configured to cause the data processing apparatus to execute a process for compressing a binary image representing a document containing text regions, the process comprising:

(a) segmenting the text regions into a plurality of symbol images, each symbol image representing a symbol of text, each symbol image being bound by a bounding box having a location and a size;

(b) classifying each symbol image obtained in step (a) into one of a plurality of classes, each class being represented by a template image and a class index, including, for each symbol image being classified;

(b1) comparing the symbol image with each template image to determine whether they match each other, including comparing a plurality of features of the symbol image with the corresponding plurality of features of the template image, the plurality of features including density statistics features, side profile features, topology statistics features and shape features;

(b2) if a match is found in step (b1), recording the class index corresponding to the matched template in association with the symbol image being classified; and

(b3) if no match is found in step (b1), adding a new class to the plurality of classes, by using the image of the symbol image being classified as the template image of the new class and assigning a class index to the new class, and recording the class index in association with the symbol image being classified;

(c) resizing the template image of each class to a final size; and

(d) storing, as compressed image data, the resized template image for each of the plurality of classes along with its class index, the bounding box location and size for each symbol image obtained in step (a), and the class index for each symbol image obtained in step (b2) or (b3).

12. The computer program product of claim 11, wherein step (a) includes segmenting the text regions into a plurality of lines, segmenting the lines into a plurality of words, and segmenting the words into the plurality of symbol images.

13. The computer program product of claim 11, wherein the density statistics features include zoning profile features, wherein the shape features include low-order moment features, and

wherein step (b1) includes, for each template image being compared:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the symbol image that are different from corresponding features of the template image; and

wherein the symbol image and the template image are determined to match each other if;

the first difference number is smaller than or equal to a first threshold value, the second difference number is smaller than or equal to a second threshold value, the third difference number is smaller than or equal to a third threshold value, and the fourth difference number is smaller than or equal to a fourth threshold value.

14. The computer program product of claim 11, wherein the density statistics features include zoning profile features, wherein the shape features include low-order moment features, and

wherein step (b1) includes, for each template image being compared:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the symbol image that are different from corresponding features of the template image; and

wherein the symbol image and the template image are determined to match each other if a sum of the first through fourth difference numbers is smaller than or equal to a threshold value.

15. The computer program product of claim 11, wherein the process further comprises, after step (b) and before step (c):

merging a first one of the plurality of classes into a second one of the plurality of classes if a total number of symbol images classified into the first class in step (b) is below a threshold and a measure of confidence of classification for the first class is below another threshold.

16. The computer program product of claim 11, wherein in step (c), the final sizes for at least some template images are different from each other, wherein step (c) includes:

(c1) calculating a similarity measure between each template image with each other template image;

(c2) determining a final size for each template image based on the calculated similarity measure with other template images; and

(c3) resizing each template image to the final size determined in step (c2).

17. The computer program product of claim 16, wherein step (c1) includes, for each pair of template images:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the pair of template image that are different from each other; and

summing the first through fourth difference numbers as the similarity measure.

18. A computer program product comprising a computer usable non-transitory medium having a computer readable program code embedded therein for controlling a data processing apparatus, the computer readable program code configured to cause the data processing apparatus to execute a process for compressing a binary image representing a document containing text regions, the process comprising:

(a) segmenting the text regions into a plurality of symbol images, each symbol image representing a symbol of text, each symbol image being bound by a bounding box having a location and a size;

(b) classifying each symbol image obtained in step (a) into one of a plurality of classes, each class being represented by a template image and a class index, including, for each symbol image being classified;

(b1) comparing the symbol image with each template image to determine whether they match each other;

(b2) if a match is found in step (b1), recording the class index corresponding to the matched template in association with the symbol image being classified; and

(b3) if no match is found in step (b1), adding a new class to the plurality of classes, by using the image of the symbol image being classified as the template image of the new class and assigning a class index to the new class, and recording the class index in association with the symbol image being classified;

(c) resizing the template image of each class to a final size, wherein the final sizes for at least some template images are different from each other, including;

(c1) calculating a similarity measure between each template image with each other template image;

(c2) determining a final size for each template image based on the calculated similarity measure with other template images; and

(c3) resizing each template image to the final size determined in step (c2); and

(d) storing, as compressed image data, the resized template image for each of the plurality of classes along with its class index, the bounding box location and size for each symbol image obtained in step (a), and the class index for each symbol image obtained in step (b2) or (b3).

19. The computer program product of claim 18, wherein step (c1) includes, for each pair of template images:

calculating a first, a second, a third and a fourth difference number representing, respectively, a number of zoning profile features, a number of side profile features, a number of topology statistics features, and a number of low-order moment features of the pair of template image that are different from each other; and

summing the first through fourth difference numbers as the similarity measure.

20. The computer program product of claim 18, wherein the process further comprises, after step (b) and before step (c):

merging a first one of the plurality of classes into a second one of the plurality of classes if a total number of symbol images classified into the first class in step (b) is below a threshold and a measure of confidence of classification for the first class is below another threshold.