OCR classification based on transition ground data

• US 5,781,659 A
• Filed: 02/04/1997
• Issued: 07/14/1998
• Est. Priority Date: 08/26/1993
• Status: Expired due to Term

• Current Assignee: Canon Inc.
• Original Assignee: Canon Inc.
• Inventors: Avi-Itzhak, Hadar, Melen, Roger D.
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): DEL ROSSO, GERARD DOMNICK
• Application Number:
  US08/794,964
• Time in Patent Office: 525 Days
• Field of Search: 382/209, 382/217, 382/218, 382/224, 382/226, 382/200, 382/194, 382/228, 382/187, 382/155
• US Class Current: 382/187
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06V 10/751 Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching

Abstract

An OCR system 10 classifies an input image vector of an unclassified symbol with respect to a library 14T of template image vectors of pre-classified characters. Each template vector is in the form of a sequence of elements representing the image intensity level of a pixel within the character defined by that template vector. Each template element is part of the image background, foreground, or transition ground between the background and foreground. Each input vector, like the template vectors, is also in the form of a sequence of elements. However, in the input vector, each element represents the sum or an image intensity level signal component defining the symbol within the image of the unclassified symbol plus a greyscale noise component. Each input element is also part of the background, foreground, or transition ground. The input vector and at least one of the template vectors are entered into a classifier device 18. The input vector is classified based on the backgrounds, foregrounds, and transition grounds. The presence of transition ground in the input vector and the template vector produces a robust classification response with a more uniform correlation coefficient between repeated classifications of the same input symbol. The classifier device may be a distance function classifier or a neural network classifier.

Claims

1. An OCR (optical character recognition) method of classifying an input image vector of an unclassified whole character with respect to a library of template image vectors of pre-classified characters, said method comprising the steps of:

providing a library of template image vectors of pre-classified characters each in the form of a sequence of template elements, each template element having a numerical value representing the image intensity level of a corresponding element within the pre-classified character defined by the template vector, said template elements including image background elements, image foreground elements, and image transition ground elements that are situated between image background elements and image foreground elements and are caused by a progressive change from background intensity level to foreground intensity level;

providing an input image vector of an unclassified whole character in the form of a sequence of input elements, each input element having a numerical value representing an image intensity level signal component corresponding to an element within the unclassified character, said input elements being from the group of elements comprising image background elements, image foreground elements, and image ground elements that are situated between image background elements and image foreground elements and are caused by a progressive change from background intensity level to foreground intensity level;

entering the input vector and at least one of the template vectors into a classifier means; and

classifying the input vector with respect to the entered template vectors by comparing numerical values of elements of the input vector with numerical values of elements of entered template vectors.

2. The OCR method of claim 1, wherein the classifier means is a distance function.classifier for comparing the input vector to each of the entered template vectors on an element by element basis and summing the element comparisons to classify the input vector.

3. The OCR method of claim 1, wherein the classifier means is a neural network classifier which develops a learned data base therein in response to the entered template vectors to classify the input vector.

4. The OCR method of claim 1, further comprising, before the classifying step, the additional steps of:

identifying the elements of the input vector having a numerical value within a predetermined background clamping range of numerical values; and

clamping the numerical value of each of the identified elements to a predetermined background clamping numerical value.

5. The OCR method of claim 4, wherein the predetermined clamping level is within the predetermined clamping range.

6. The OCR method of claim 4, wherein the background clamping level is a general background signal level determined by the levels of the signal components of the background elements of the input vector, the portion of the numerical value of each background element below the general background signal level defines a hyper background noise sub-component of a greyscale noise component of that element, the background clamping range is a hyper background clamping range determined by the hyper background noise sub-components, and clamping the numerical value of elements within the hyper background range eliminates hyper background noise by clamping the numerical value of elements within the hyper background range to the general background signal level.

7. The OCR method of claim 6, wherein the portion of the numerical value of each element near the general background signal level defines a near background noise sub-component of the greyscale noise component of that element, the background clamping range is the hyper background clamping range plus a near background clamping range determined by the near background noise sub-components, and the clamping of the hyper background range together with the near background range eliminates hyper background noise plus near background noise by clamping elements within the hyper background range and the near background range to the general background signal level.

8. The OCR method of claim 4, wherein the background clamping level is a general background signal level determined by the levels of the signal component of the background elements of the input vector, the portion of the numerical value of each element near the general background signal level defines a near background noise sub-component of a greyscale noise component of that element, the background clamping range is a near background clamping range determined by the near background noise sub-components, and clamping elements within the near background range eliminates near background noise by clamping the near background range to the general background signal level.

9. The OCR method of claim 4, wherein the intensity of each element of the signal components of the input vector is between a minimum limit value and a maximum limit value, with the intensities of the background elements closer to one of the limit values than to the other one of the limit values, and the intensities of the foreground elements closer to the other one of the limit values than to said one of the limit values.

10. The OCR method of claim 9, wherein the background clamping level is a general background signal level determined by the intensity level of the signal component of the background elements of the input vector.

11. The OCR method of claim 9, wherein the background clamping level has the numerical value of 0.00.

12. The OCR method of claim 9, wherein the background clamping level is a predetermined number of levels away from the minimum limit value.

13. The OCR method of claim 1, further comprising, before the classifying step, the additional steps of:

identifying the elements of the input vector having a numerical value within a predetermined foreground clamping range of numerical values; and

clamping the numerical value of each of the identified elements to a predetermined foreground clamping numerical value.

14. The OCR method of claim 13, wherein the foreground clamping level is a general foreground signal level determined by the level signal component of the numerical values of the foreground elements of the input vector, the portion of the numerical value of each foreground element above the general foreground signal level defines a hyper foreground noise sub-component of a greyscale noise component of that element, the foreground clamping range is a hyper foreground clamping range determined by the hyper foreground noise sub-components, and clamping elements within the hyper foreground range eliminates hyper foreground noise by clamping elements within the hyper foreground range to the general foreground signal level.

15. The OCR method of claim 14, wherein the portion of the numerical value of each element near the general foreground signal level defines a near foreground noise sub-component of the greyscale noise component of that element, the foreground clamping range is the hyper foreground clamping range plus a near foreground clamping range determined by the near foreground noise sub-components, and the clamping of elements within the hyper foreground range together with the near foreground range eliminates hyper foreground noise plus near foreground noise by clamping elements within the hyper foreground range and the near foreground clamping range to the general foreground signal level.

16. The OCR method of claim 13, wherein the foreground clamping level is a general foreground signal level determined by the level signal component of the numerical values of the foreground elements of the input vector, the portion of the numerical value of each element near the general foreground signal level defines a near foreground noise sub-component of a greyscale noise component of that element, the foreground clamping range is a near foreground clamping range determined by the near foreground noise sub-components, and clamping elements within the near foreground range eliminates near foreground noise by clamping elements within the near foreground range to the general foreground signal level.

17. The OCR method of claim 13, wherein the intensity of each element of the level signal components of the input vector is between a minimum limit value and a maximum limit value, with the intensities of the background elements closer to one of the limit values, and the intensities of the foreground elements closer to the other one of the limit values.

18. The OCR method of claim 17, wherein the foreground clamping level is a general foreground signal level determined by the intensity level of the signal component of the foreground elements of the input vector.

19. The OCR method of claim 17, wherein the foreground clamping level is 1.00.

20. The OCR method of claim 17, wherein the foreground clamping level is a predetermined number of levels away from the maximum limit value.

21. The OCR method of claim 1, further comprising, before the classifying step, the additional steps of:

identifying the elements of the input vector having a numerical value within a predetermined background clamping range of numerical values;

clamping the numerical value of each of the identified elements to a predetermined background clamping numerical value within the predetermined background clamping range;

identifying the elements of the input vector having a numerical value within a predetermined foreground clamping range of numerical values; and

clamping the numerical value of each of the identified elements to a predetermined foreground clamping numerical value within the predetermined foreground clamping range.

22. The OCR method of claim 21, wherein the background clamping level is a general background signal level determined by the level signal component of the numerical values of the background elements of the input vector, the portion of the numerical value of each background element below the general background signal level defines a hyper background nose sub-component of a greyscale noise component of that element, the background clamping range is a hyper background clamping range determined by the hyper background noise sub-components, and clamping elements within the hyper background range eliminates hyper background noise by clamping elements within the hyper background range to the general background signal level.

23. The OCR method of claim 22, wherein the portion of the numerical value of each element near the general background signal level defines a near background noise sub-component of the greyscale noise component of that element, the background clamping range is the hyper background clamping range plus a near background clamping range determined by the near background noise sub-components, and the clamping of the hyper background range together with the near background range eliminates hyper background noise plus near background noise by clamping the elements within the hyper background range and the near background clamping range to the general background signal level.

24. The OCR method of claim 21, wherein the background clamping level is a general background signal level determined by the level signal component of the numerical values of the background elements of the input vector, the portion of the numerical value of each element near the general background signal level defines a near background noise sub-component of a greyscale noise component of that element, the background clamping range is a near background clamping range determined by the near background noise sub-components, and clamping elements within the near background range eliminates near background noise by clamping elements within the near background range to the general background signal level.

25. The OCR method of claim 21, wherein the foreground clamping level is a general foreground signal level determined by the level signal component of the numerical values of the foreground elements of the input vector, the portion of the numerical value of each foreground element above the general foreground signal level defines a hyper foreground noise sub-component of a greyscale noise component of that element, the foreground clamping range is a hyper foreground clamping range determined by the hyper foreground noise sub-components, and clamping elements within the hyper foreground range eliminates hyper foreground noise by clamping elements within the hyper foreground range to the general foreground signal level.

26. The OCR method of claim 25, wherein the portion of the numerical value of each element near the general foreground signal level defines a near foreground noise sub-component of the greyscale noise component of that element, the foreground clamping range is the hyper foreground clamping range plus a near foreground clamping range determined by the near foreground noise sub-components, and the clamping of the hyper foreground range together with the near foreground range eliminates hyper foreground noise plus near foreground noise by clamping elements within the hyper foreground range and the near foreground clamping range to the general foreground signal level.

27. The OCR method of claim 21, wherein the foreground clamping level is a general foreground signal level determined by the level signal component of the numerical values of the foreground elements of the input vector, the portion of the numerical value of each element near the general foreground signal level defines a near foreground noise sub-component of a greyscale noise component of that element, the foreground clamping range is a near foreground clamping range determined by the near foreground noise sub-components, and clamping elements within the near foreground range eliminates near foreground noise by clamping elements within the near foreground range to the general foreground signal level.

28. The OCR method of claim 1, further comprising, before the classifying step, the additional steps of:

identifying the elements of the input vector having a numerical value within a predetermined input clamping range of numerical values;

clamping the numerical value of each of the identified input elements to a predetermined input clamping numerical value;

identifying the elements of the template vectors having a numerical value within a predetermined template clamping range of numerical values; and

clamping the numerical value of each of the identified template elements to a predetermined template clamping numerical value.

29. The OCR method of claim 28, wherein the predetermined template clamping range is the same as the predetermined input clamping range, and the predetermined template clamping level is the same as the predetermined input clamping level.

30. The OCR method of claim 1, further comprising, before the classifying step, the additional step of:

sorting the elements of the input vector in order of the numerical values thereof.

31. The OCR method of claim 30, further comprising, after the input vector sorting step, the additional step of:

sorting the elements of the template vector in order of the numerical values thereof.

32. The OCR method of claim 1, wherein numerical values of the background elements of the input vector determine a general background input signal level, numerical values of the foreground elements of the input vector determine a general foreground input signal level, numerical values of the background elements of the entered template vector determine a general background template signal level, and numerical values of the foreground elements of the entered template vector determine a general foreground template signal level.

33. The OCR method of claim 32, further comprising, before the classifying step, the additional steps of:

setting the general background signal level of the input vector and the entered template vector to the same general background signal level; and

setting the general foreground signal level of the input vector and the entered template vector to the same general foreground signal level.

34. The OCR method of claim 32, further comprising, before the classifying step, the additional steps of:

setting the general background signal level of the input vector to the general background signal level of the entered template vector; and

setting the general foreground signal level of the input vector to the general foreground signal level of the entered template vector.

35. The method of claim 1, wherein the input elements include image background elements, image foreground elements, and transition around elements.

36. The OCR method of claim 1, wherein an input image is input-scanned by a scanner, and said input image vector is extracted from said input image.