System and method for increasing the accuracy of optical character recognition (OCR)

• US 20110103688A1
• Filed: 11/02/2009
• Published: 05/05/2011
• Est. Priority Date: 11/02/2009
• Status: Active Grant

• Current Assignee: Hyland Switzerland, Sarl (HSI Holdings II Incorporated)
• Original Assignee: Lexmark Enterprise Software LLC (Tungsten Automation Corp.)
• Inventors: Hausmann, Johannes, Meier, Ralph, Urbschat, Harry, Wanschura, Thorsten
• Granted Patent: US 9,152,883 B2
• Time in Patent Office: Days
• US Class Current: 382/182
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/12 Detection or correction of errors, e.g. by rescanning the pattern
• CPC Class Codes: G06V 30/1914 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries, e.g. user dictionaries

Abstract

A system and/or method for increasing the accuracy of optical character recognition (OCR) for at least one item, comprising: obtaining OCR results of OCR scanning from at least one OCR module; creating at least one OCR seed using at least a portion of the OCR results; creating at least one OCR learn set using at least a portion of the OCR seed; and applying the OCR learn set to the at least one item to obtain additional optical character recognition (OCR) results.

Claims

1. A method for increasing the accuracy of optical character recognition (OCR) for at least one item, comprising:

obtaining OCR results of OCR scanning from at least one OCR module;

creating at least one OCR seed using at least a portion of the OCR results;

creating at least one OCR learn set using at least a portion of the OCR seed; and

applying the OCR learn set to the at least one item to obtain additional optical character recognition (OCR) results.

2. The method of claim 1, wherein image statistics are deducted based on the at least one OCR learn set.

3. The method of claim 1, wherein the at least one item comprises:

at least one page;

at least one document;

at least one part of at least one document;

at least one zone;

at least one paragraph;

at least one line;

at least one word;

or

at least one character;

or

any combination thereof.

4. The method of claim 1, wherein the at least one OCR learn set is used on different items.

5. The method of claim 1, wherein different OCR modules and/or different parameter sets and/or different preprocessing steps are used.

6. The method of claim 1, wherein the additional OCR results take into account:

font size;

font type;

or

font properties;

or

any combination thereof.

7. The method of claim 1, wherein whitespace characteristics are taken into account in creating the at least one OCR learn set.

8. The method of claim 1, wherein the method is preformed iteratively to allow for a refinement of the method.

9. The method of claim 1, wherein the at least one OCR seed is created by compiling imagelets corresponding to each character identified in the OCR results;

and the imagelets can be stored imagelets and/or new imagelets.

10. The method of claim 9, wherein the OCR learn set includes manually input examples.

11. The method of claim 15, wherein statistical measurements and/or cluster analysis is used to select similar imagelets.

12. The method of claim 11, wherein the statistical measurements and/or the cluster analysis comprises:

at least one Bayes classifier;

at least one neural networks classifier;

at least one support vector classifier;

or

any combination thereof.

13. The method of claim 9, wherein stored imagelets are used to increase the at least one OCR learn set.

14. The method of claim 9, wherein the at least one OCR learn set and related statistics are used to derive at least one image filtering technique.

15. The method of claim 9, wherein the at least one OCR seed is cleaned by selecting similar imagelets for each character identified in the OCR results.

16. The method of claim 1, wherein the OCR learn set is created by obtaining information related to the image representation for each character and the variability of the similar imagelets from the image representation for each character in the at least one OCR cleaned seed.

17. The method of claim 16, wherein the image representation is a bitmap.

18. The method of claim 17, wherein the bitmap is grayscale or color.

19. The method of claim 9, further comprising:

comparing the at least one OCR learn set to each imagelet to create at least one mismatch distribution of the at least one OCR learn set compared to each imagelet; and

generating the OCR additional results based on the at least one mismatch distribution.

20. A system for increasing the accuracy of optical character recognition (OCR) for at least one item, comprising:

at least one processor, wherein the at least one processor is configured to perform;

obtaining OCR results of OCR scanning from at least one OCR module;

creating at least one OCR seed using at least a portion of the OCR results;

creating at least one OCR learn set using at least a portion of the OCR seed; and

applying the OCR learn set to the at least one item to obtain additional optical character recognition (OCR) results.

21. The system of claim 20, wherein image statistics are deducted based on the at least one OCR learn set.

22. The system of claim 20, wherein the at least one item comprises:

at least one page;

at least one document;

at least one part of at least one document;

at least one zone;

at least one paragraph;

at least one line;

at least one word;

or

at least one character;

or

any combination thereof.

23. The system of claim 20, wherein the at least one OCR learn set is used on different items.

24. The system of claim 20, wherein different OCR modules and/or different parameter sets and/or different preprocessing steps are used.

25. The system of claim 20, wherein the additional OCR results take into account:

font size;

font type;

or

font properties;

or

any combination thereof.

26. The system of claim 20, wherein whitespace characteristics are taken into account in creating the at least one OCR learn set.

27. The system of claim 20, wherein the method is preformed iteratively to allow for a refinement of the method.

28. The system of claim 20, wherein the at least one OCR seed is created by compiling imagelets corresponding to each character identified in the OCR results.

29. The system of claim 28, wherein the at least one OCR learn set comprises stored imagelets and/or new imagelets.

30. The system of claim 34, wherein statistical measurements and/or cluster analysis is used to select similar imagelets.

31. The system of claim 30, wherein the statistical measurements and/or the cluster analysis comprises:

at least one Bayes classifier;

at least one neural networks classifier;

at least one support vector classifier;

or

any combination thereof.

32. The system of claim 28, wherein stored imagelets are used to increase the at least one OCR learn set.

33. The system of claim 28, wherein the at least one OCR learn set and related statistics are used to derive at least one image filtering technique.

34. The system of claim 28, wherein the at least one OCR seed is cleaned by selecting similar imagelets for each character identified in the OCR results.

35. The system of claim 20, wherein the OCR learn set is created by obtaining information related to the image representation for each character and the variability of the similar imagelets from the image representation for each character in the at least one OCR cleaned seed.

36. The method of claim 35, wherein the image representation is a bitmap.

37. The method of claim 36, wherein the bitmap is grayscale or color.

38. The method of claim 28, further comprising:

comparing the at least one OCR learn set to each imagelet to create at least one mismatch distribution of the at least one OCR learn set compared to each imagelet; and

generating the OCR additional results based on the at least one mismatch distribution.