ACCURACY OF RECOGNITION BY MEANS OF A COMBINATION OF CLASSIFIERS

• US 20110274345A1
• Filed: 05/06/2010
• Published: 11/10/2011
• Est. Priority Date: 05/06/2010
• Status: Active Grant

• Current Assignee: ABBYY Development LLC
• Original Assignee: ABBYY Software
• Inventors: Tuganbaev, Diar
• Granted Patent: US 8,660,371 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06V 30/10 Character recognition

Abstract

In one embodiment, there is provided a method for an Optical Character Recognition (OCR) system. The method comprises: recognizing an input character based on a plurality of classifiers, wherein each classifier generates an output by comparing the input character with a plurality of trained patterns; grouping the plurality of classifiers based on a classifier grouping criterion; and combining the output of each of the plurality of classifiers based on the grouping.

Claims

1. A method for an Optical Character Recognition (OCR) system, the method comprising:

recognizing an input character based on a plurality of classifiers, wherein each classifier generates an output by comparing the input character with a plurality of trained patterns;

grouping the plurality of classifiers based on a classifier grouping criterion; and

combining the output of each of the plurality of classifiers based on the grouping.

2. The method of claim 1, wherein the classifier grouping criterion includes a measure of a similarity between the classifiers.

3. The method of claim 1, wherein combining the output of classifiers comprises combining a weight output by each classifier of at least one same type group into at least one interim weight, as part of a first stage.

4. The method of claim 3, wherein the interim weight is a final weight in a case of there being only one type group at the first stage and no second stage.

5. The method of claim 3, wherein the combining comprises a second stage in which each interim weight is combined with one of a further weight output by a classifier of a second type and another interim weight.

6. The method of claim 5, wherein the second stage is performed multiple times in which case for each additional time, the interim weight is the output of a last performance of the second stage.

7. The method of claim 3, wherein the weights are combined in accordance with a formula for a weighted arithmetic mean.

8. The method of claim 7, wherein weight coefficients for the formula of a weighted arithmetic mean are selected separately for each character on a basis of a preliminary training process.

9. The method of claim 8, wherein the preliminary training process comprises selecting the weight coefficients which best distinguish images of the character (for which weight coefficients are being selected) from other images.

10. The method of claim 9, wherein the weight coefficients are selected on a basis of searching an extremum of a function, which distinguishes the images as a hyperplane.

11. The method of claim 10, wherein the function is expressed as

12. The method of claim 5, wherein an output of each performance of the second stage is generated based on a table of final weights that maps weight pairs (x,x_n+1) to a final weight w.

13. The method of claim 12, wherein correspondences (x,x_n+1)−

w are selected based on a technique that maintains a percentage of right and wrong images associated with each weight pair;

where x is the interim weight, and x_n+1 the weight output by the classifier of the second type.

14. The method of claim 4, wherein the classifier of the second type is a structure classifier.

15. An Optical Character Recognition (OCR) system comprising:

at least one memory; and

at least a processor controlled by stored programmed instructions in the at least one memory to perform functions of

recognizing an input character based on a plurality of classifiers, wherein each classifier generates an output by comparing the input character with a plurality of trained patterns;

grouping the plurality of classifiers based on a classifier grouping criterion; and

combining the output of each of the plurality of classifiers based on the grouping.

16. The OCR system of claim 15, wherein the classifier grouping criterion includes a measure of a similarity between the classifiers.

17. The OCR system of claim 15, wherein combining the output of classifiers comprises combining a weight output by each classifier of at least one same type group into at least one interim weight, as part of a first stage.

18. The OCR system of claim 17, wherein the interim weight is a final weight in a case of there being only one type group at the first stage and no second stage.

19. The OCR system of claim 17, wherein the combining comprises a second stage in which each interim weight is combined with one of a further weight output by a classifier of a second type and another interim weight.

20. The OCR system of claim 19, wherein the second stage is performed multiple times in which case for each additional time, the interim weight is the output of a last performance of the second stage.

21. The OCR system of claim 17, wherein the weights are combined in accordance with a formula for a weighted arithmetic mean.

22. The OCR system of claim 21, wherein weight coefficients for the formula of a weighted arithmetic mean are selected separately for each character on a basis of a preliminary training process.

23. The OCR system of claim 22, wherein the preliminary training process comprises selecting the weight coefficients which best distinguish images of the character (for which weight coefficients are being selected) from other images.

24. The OCR system of claim 23, wherein the weight coefficients are selected on a basis of searching an extremum of a function, which distinguishes the images as a hyperplane.

25. The OCR system of claim 24, wherein the function is expressed as

26. The OCR system of claim 19, wherein an output of each performance of the second stage is generated based on a table of final weights that maps weight pairs (x,x_n+1) to a final weight w.

27. The OCR system of claim 26, wherein correspondences (x,x_n+1)−

w are selected based on a technique that maintains a percentage of right and wrong images associated with each weight pair;

where x is the interim weight, and x_n+1 is the weight output by the classifier of the second type.

28. The OCR system of claim 18, wherein the classifier of the second type is a structure classifier.

29. A computer readable medium having stored thereon a sequence of instructions which when executed by a computer, causes the computer to perform a method for an Optical Character Recognition (OCR) system, the method comprising:

recognizing an input character based on a plurality of classifiers, wherein each classifier generates an output by comparing the input character with a plurality of trained patterns;

grouping the plurality of classifiers based on a classifier grouping criterion; and

combining the output of each of the plurality of classifiers based on the grouping.

30. The computer readable medium of claim 29, wherein the classifier grouping criterion includes a measure of a similarity between the classifiers.

31. The computer readable medium of claim 29, wherein combining the output of classifiers comprises combining a weight output by each classifier of at least one same type group into at least one interim weight, as part of a first stage.

32. The computer readable medium of claim 31, wherein the interim weight is a final weight in a case of there being only one type group at the first stage and no second stage.

33. The computer readable medium of claim 31, wherein the combining comprises a second stage in which each interim weight is combined with one of a further weight output by a classifier of a second type and another interim weight.

34. The computer readable medium of claim 33, wherein the second stage is performed multiple times in which case for each additional time, the interim weight is the output of a last performance of the second stage.

35. The computer readable medium of claim 31, wherein the weights are combined in accordance with a formula for a weighted arithmetic mean.

36. The computer readable medium of claim 35, wherein weight coefficients for the formula of a weighted arithmetic mean are selected separately for each character on a basis of a preliminary training process.

37. The computer readable medium of claim 36, wherein the preliminary training process comprises selecting the weight coefficients which best distinguish images of the character (for which weight coefficients are being selected) from other images.

38. The computer readable medium of claim 37, wherein the weight coefficients are selected on a basis of searching an extremum of a function, which distinguishes the images as a hyperplane.

39. The computer readable medium of claim 38, wherein the function is expressed as

40. The computer readable medium of claim 33, wherein an output of each performance of the second stage is generated based on a table of final weights that maps weight pairs (x,x_n+1) to a final weight w.

41. The computer readable medium of claim 40, wherein correspondences (x,x_n+1)−

w are selected based on a technique that maintains a percentage of right and wrong images associated with each weight pair;

where x is the interim weight, and x_n+1 is the weight output by the classifier of the second type.

42. The computer readable medium of claim 32, wherein the classifier of the second type is a structure classifier.