ADAPTIVE PARTIAL CHARACTER RECOGNITION

• US 20100166318A1
• Filed: 12/30/2008
• Published: 07/01/2010
• Est. Priority Date: 12/30/2008
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Ben-Horesh, Ami, Walach, Eugeniusz, Geva, Amir
• Granted Patent: US 8,306,327 B2
• Time in Patent Office: Days
• US Class Current: 382/195
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/155 Removing patterns interfering with the pattern to be recognised, such as ruled lines or underlines
• CPC Class Codes: G06V 30/1902 Shifting or otherwise transforming the patterns to accommodate for positional errors
• CPC Class Codes: G06V 30/248 involving plural approaches, e.g. verification by template match; Resolving confusion among similar patterns, e.g. "O" versus "Q" G06V30/242 takes precedence

Abstract

A method and system for recognizing a character affected by a noise or an obstruction is disclosed. After receiving an image with characters, a character being affected by a noise or an obstruction is determined. Then, areas in the character where the noise or obstruction affected are precisely located. Templates representing every possible character in the image are updated by removing equivalent areas to the areas in the character being affected by the noise or obstruction. Then, the character is classified in a template among the updated templates by finding the template having the highest number of matching pixels with the character.

Claims

1. A system for recognizing a character with a noise or an obstruction in an image, the image including a plurality of characters, comprising:

means for detecting the noise or the obstruction in the character;

means for providing one or more templates of the plurality of characters;

means for updating the one or more templates of the plurality of characters based on the detected noise or the detected obstruction; and

means for classifying the character in one of the updated templates of the plurality of characters, the character being recognized as a letter, number or symbol that the template, which the character is classified in, represents.

2. The system according to claim 1, further comprising:

means for calculating a stroke feature of unobstructed areas of the character, the unobstructed areas of the character not having any effect from a noise or an obstruction; and

means for calculating statistics of adjacent characters of the character, the adjacent characters not having any effect from a noise or an obstruction.

3. The system according to claim 2, wherein the means for detecting performs finding at least one area of the character in which a stroke feature does not conform to the calculated stroke feature of the unobstructed areas or in which statistics does not conform to the calculated statistics of the adjacent characters.

4. The system according to claim 3, wherein the means for updating comprises:

means for removing the at least one area in the one or more templates; and

means for removing the at least one area in the character.

5. The system according to claim 3, wherein the means for classifying comprises:

means for counting the number of matching first-type pixels in a template and the character;

means for counting the number of matching second-type pixels in the template and the character;

means for setting pixels in the at least one area in the character as don'"'"'t care pixels;

means for counting the number of don'"'"'t care pixels in the character;

means for calculating a value of (the number of matching first-type pixels+the number of matching second-type pixels−

the number of don'"'"'t care pixels) for each pair of each template and the character;

means for selecting a pair of a template and the character that has a highest calculated value among the calculated values of pairs of each template and the character; and

means for identifying the character as the letter, the number or the symbol that the template in the selected pair represents.

6. The system according to claim 2, wherein the stroke feature is one or more of:

a stroke width, a stroke length, a stroke size, a stroke duration, a stroke orientation, a stroke shape and a stroke pattern.

7. The system according to claim 2, wherein the statistics is one or more of

a font height, an ascent value and a descent value.

8. The system according to claim 2, further comprising:

means for dividing the character into parts; and

means for analyzing each part to determine whether the part is affected by the noise or the obstruction.

9. The system according to claim 8, further comprising:

means for further dividing a part into subparts, if the part is affected by the noise or the obstruction;

means for finding at least one area in a subpart, the at least one area does not conform to the calculated stroke feature.

10. The system according to claim 1, wherein the means for classifying performs a comparison between the updated templates and the character to recognize the character.

11. The system according to claim 10, wherein the means for classifying performs the comparison based on one or more of:

correlating each of the updated templates with the character, finding difference between each of the updated templates and the character, and utilizing a trained neural network to identify the character.

12. A method for recognizing a character with a noise or an obstruction in an image, the image including a plurality of characters, comprising:

detecting the noise or the obstruction in the character;

providing one or more templates of the plurality of characters;

updating the one or more templates of the plurality of characters based on the detected noise or the detected obstruction; and

classifying the character in one of the updated templates of the plurality of characters, the character being recognized as a letter, number or symbol that the template, which the character is classified in, represents.

13. The method according to claim 12, further comprising:

calculating a stroke feature of unobstructed areas of the character, the unobstructed areas of the character not having any effect from a noise or an obstruction; and

calculating statistics of adjacent characters of the character, the adjacent characters not having any effect from a noise or an obstruction.

14. The method according to claim 13, wherein the detecting performs finding at least one area of the character in which a stroke feature does not conform to the calculated stroke feature of the unobstructed areas or in which statistics does not conform to the calculated statistics of the adjacent characters.

15. The method according to claim 14, wherein the updating comprises:

removing the at least one area in the one or more templates; and

removing the at least one area in the character.

16. The method according to claim 14, wherein the classifying comprises:

counting the number of matching first-type pixels in a template and the character;

counting the number of matching second-type pixels in the template and the character;

setting pixels in the at least one area in the character as don'"'"'t care pixels;

counting the number of don'"'"'t care pixels in the character;

calculating a value of (the number of matching first-type pixels+the number of matching second-type pixels−

the number of don'"'"'t care pixels) for each pair of each template and the character;

selecting a pair of a template and the character that has a highest calculated value among the calculated values of pairs of each template and the character; and

identifying the character as the letter, the number or the symbol that the template in the selected pair represents.

17. The method according to claim 13, wherein the stroke feature is one or more of:

a stroke width, a stroke length, a stroke size, a stroke duration, a stroke orientation, a stroke shape and a stroke pattern.

18. The system according to claim 13, wherein the statistics is one or more of:

a font height, an ascent value and a descent value.

19. The method according to claim 13, further comprising:

dividing the character into parts; and

analyzing each part to determine whether the part is affected by the noise or the obstruction.

20. The method according to claim 19, further comprising:

further dividing a part into subparts, if the part is affected by the noise or the obstruction;

finding at least one area in a subpart, the at least one area does not conform to the calculated stroke feature.

21. The method according to claim 12, wherein the classifying performs a comparison between the updated templates and the character to recognize the character.

22. The method according to claim 21, wherein the classifying performs the comparison based on one or more of:

correlating each of the updated templates with the character, finding difference between each of the updated templates and the character, and utilizing a trained neural network to identify the character.

23. A computer readable medium embodying computer program instructions being executed by a processor for causing a computer to perform method steps for recognizing a character with a noise or an obstruction in an image, the image including a plurality of characters, said method steps comprising the steps of claim 12.

24. A method of deploying a computer program product including a program of instructions in a computer readable medium for recognizing a character with a noise or an obstruction in an image, the image including a plurality of characters, wherein, when the program of instructions is executed by a processor, the computer program product performs the steps of claim 12.