Method And Apparatus For Recognizing Characters In A Document Image

• US 20080310721A1
• Filed: 06/14/2007
• Published: 12/18/2008
• Est. Priority Date: 06/14/2007
• Status: Abandoned Application

• Current Assignee: Seiko Epson Corporation (Seiko Group)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Zhou, Hui, Achong, Jeffrey Matthew, Vafi, Narges, Yang, John Jinhwan
• Application Number:
  US11/763,000
• Publication Number:
  US 20080310721A1
• Time in Patent Office: Days
• US Class Current: 382/182
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1475 Inclination or skew detection or correction of characters or of image to be recognised
• CPC Class Codes: G06V 30/162 Quantising the image signal
• CPC Class Codes: G06V 30/164 Noise filtering
• CPC Class Codes: G06V 30/182 by coding the contour of the pattern

Abstract

A method of recognizing characters in a document image comprises examining the intensity of pixels in the document image and identifying a peak intensity deemed to represent foreground in the document image. A threshold level for distinguishing the foreground from background in the document image as a function of the identified peak intensity is determined. The document image is thresholded using the threshold level to identify the foreground. Character recognition is performed on the foreground of the document image.

Claims

1. A method of recognizing characters in a document image, comprising:

examining the intentisity of pixels in said document image;

identifying a peak intensity deemed to represent foreground in said document image;

determining a threshold level for distinguishing foreground from background as a function of said identified peak intensity;

thresholding said document image using said threshold level to identify said foreground; and

performing character recognition on said identified foreground.

2. The method of claim 1, further comprising:

examining the intensity of said pixels and identifying a valley intensity following said peak intensity, wherein during said determining said threshold level is calculated as a function of said peak intensity and said valley intensity.

3. The method of claim 2, wherein said threshold level is set to said valley intensity.

4. The method of claim 3, wherein said peak intensity is used to determine a maximum value for said threshold level, said threshold level being set to the lesser of said valley intensity and said maximum value.

5. The method of claim 1, wherein said examining comprises:

generating a pixel intensity histogram and identifying the first peak intensity therein.

6. The method of claim 5, wheren said examining further comprises:

smoothing said intensity histogram to remove intensity oscillations.

7. The method of claim 6, wherein said smoothing comprises:

applying a mean filter to said intensity histogram.

8. The method of claim 7, further comprising:

examining the intensity of said pixels and identifying a valley intensity following said first peak intensity, wherein during said determining said threshold level is calculated as a function of said first peak intensity and said valley intensity.

9. The method of claim 8, wherein said threshold level is set to said valley intensity.

10. The method of claim 9, wherein said first peak intensity is used to determine a maximum value for said threshold level, said threshold level being set to the lesser of said valley intensity and said maximum value.

11. The method of claim 1 wherein said character recognition performing comprises at least one of weighted template matching and neural network analysis to identify characters in said foreground.

12. The method of claim 10 wherein said character recognition performing comprises at least one of weighted template matching and neural network analysis to identify characters in said foreground.

13. The method of claim 5 wherein said threshold level is set to a value between the intensity of said first peak intensity and a subsequent peak intensity.

14. The method of claim 13 wherein said first peak intensity is used to determine a maximum value for said threshold.

15. The method of claim 1 wherein said character recognition performing comprises the steps of:

clustering proximate groups of pixels in said document image to form candidate characters;

comparing each candidate character to character templates representing recognizable characters and recognizing the candidate character when a match occurs; and

for each candidate character that is not recognized, performing neural network analysis to recognize the candidate character.

16. The method of claim 15 further comprising:

for each candidate character that is not recognized following neural network analysis, comparing the results of character template matching and neural network analysis to determine if the combined results, result in recognition of the candidate character.

17. The method of claim 15 further comprising:

examining each candidate character to determine if the candidate character meets a character size condition; and

performing the comparing only for each candidate character meeting said character size condition.

18. The method of claim 16 further comprising examining the candidate character to determine if the candidate character represents a zero character if the combined results of character template matching and neural network analysis do not result in the candidate character being recognized.

19. An apparatus for recognizing characters in a document image, comprising:

an image analyzer examining the intensity of pixels in said document image and identifying a peak intensity deemed to represent foreground;

a thresholder determining a threshold level for distinguishing foreground from background in said document image as a function of said identified peak intensity, and thresholding said document image using said threshold level to identify said foreground; and

a character classifier performing character recognition on said foreground of said document image.

20. An apparatus according to claim 19, wherein said image analyzer identifyies a valley intensity following said identified peak intensity, and wherein said thresholder determines said threshold level as a function of said identified peak intensity and said valley intensity.

21. An apparatus according to claim 20, wherein said image analyzer generates an intensity histogram that is examined to identify said peak intensity and valley intensity.

22. A computer-readable medium embodying a computer program for recognizing characters in a document image, said computer program comprising:

computer program code for examing the intensity of pixels in said document image;

computer program code for identifying a peak intensity deemed to represent foreground in said document image;

computer program code for determining a threshold level for distinguishing foreground from background in said document image as a function of said identified peak intensity;

computer program code for thresholding said document image using said threshold level to identify said foreground; and

computer program code for performing character recognition on said foreground of said document image.

23. A method of recognizing a candidate character in a document image, comprising:

determining edge orientations and edge magnitudes of pixels in regions encompassing pixels of said candidate character; and

analyzing said edge orientations and said edge magnitudes using a classification tool thereby to recognize said candidate character.

24. The method of claim 23, wherein said classification tool is a neural network.

25. The method according to claim 24, further comprising:

dividing the pixels forming said candidate character into regions; and

aggregating said edge orientations within said regions prior to said analyzing.

26. The method according to claim 25, further comprising:

aggregating said edge magnitudes within said regions prior to said analyzing.

27. The method of claim 26, wherein the edge orientations are determined using horizontal and vertical edge detectors.

28. An apparatus for recognizing a candidate character in a document image, comprising:

an image analyzer determining edge orientations and edge magnitudes of pixels in regions encompassing pixels of said candidate character; and

a classification tool analyzing said edge orientations and said edge magnitudes in said document image thereby to recognize characters in said document image.

29. An apparatus according to claim 28, wherein said classification tool is a neural network.

30. An apparatus according to claim 29, wherein said image analyzer divides the pixels forming said candidate characters into regions, and aggregates said edge orientations within said regions prior to processing by said neural network.

31. An apparatus according to claim 30, wherein said image analyzer aggregates said edge magnitudes within said regions prior to processing by said neural network.

32. A computer-readable medium including a computer program for recognizing a candidate character in a document image, said computer program comprising:

computer program code for determining edge orientations of pixels in windows surrounding pixels of said candidate character;

computer program code for determining edge magnitudes of pixels in windows surrounding pixels of said candidate character; and

computer program code for analyzing said edge orientations and said edge magnitudes using a classification tool thereby to recognize said candidate character.