Recognition process

• US 20030086611A1
• Filed: 04/10/2001
• Published: 05/08/2003
• Est. Priority Date: 06/09/1998
• Status: Active Grant

• Current Assignee: Apple Computer Incorporated (Apple Inc.), Laboratories For Information Technology
• Original Assignee: Laboratories For Information Technology
• Inventors: Chen, Hong, Loudon, Gareth Hugh, Tng, Tai Hou
• Granted Patent: US 6,891,971 B2
• Time in Patent Office: Days
• US Class Current: 382/186
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/15 Cutting or merging image elements, e.g. region growing, watershed or clustering-based techniques
• CPC Class Codes: G06V 30/36 Matching; Classification

Abstract

A recognition process for handwritten characters, including generating sub-segments representing at least one character stroke, the sub-segments each having a boundary enclosing the at least one stroke of the sub-segment, merging the sub-segments when the distance between centroids of the sub-segments are less than a predetermined threshold, and generating segments representing possible characters from the sub-segments.

Claims

1. A recognition process for handwritten characters, including:

generating sub-segments representing at least one character stroke, said sub-segments each having a boundary enclosing the at least one stroke of the sub-segment,

merging said sub-segments when the distance between centroids of the sub-segments are less than a predetermined threshold; and

generating segments representing possible characters from said sub-segments.

2. A recognition process as claimed in claim 1, including processing a new character stroke by including said new stroke in one of said sub-segments when said stroke is within the boundary of said one of said sub-segments and generating a new sub-segment including said new stroke when said new stroke is outside of said boundary.

3. A recognition process as claimed in claims 1 or 2, wherein generating one of said segments includes processing at least three previously generated sub-segments.

4. A recognition process as claimed in claim 3, wherein said merging step includes merging said sub-segments using a plurality of predetermined thresholds for said distance.

5. A recognition process as claimed in claim 4, wherein said thresholds represent a range of distance values.

6. A recognition process as claimed in claim 1, including merging adjacent segments which produce a merged boundary having a height to width ratio closer to one than the separate boundaries of the adjacent segments.

7. A recognition process as claimed in claim 1, 2, 4, 5 or 6, including performing character recognition on said segments to generate a segment list representing said segments and respective possible characters associated with said segments.

8. A recognition process as claimed in claim 7, including generating a time sequence representation of said possible characters from said segment list and processing said time sequence representation using a language model to generate text representing recognised characters.

9. A recognition process as claimed in claim 1, wherein said generating and merging steps are executed in the real-time as character strokes are written.

29. A computer-readable medium as claimed in claim 1, wherein the steps performed include processing a new character stroke by including said new stroke in one of said sub-segments when said stroke is within the boundary of said one of said sub-segments and generating a new sub-segment including said new stroke when said new stroke is outside of said boundary.

10. A recognition apparatus having:

means for inputting character strokes;

means for generating sub-segments representing at least one character stroke, said sub-segments each having a boundary enclosing the at least one stroke of the sub-segment,

means for merging said sub-segments when the distance between centroids of the sub-segments are less than a predetermined threshold; and

means for generating segments representing possible characters from said sub-segments.

11. A recognition apparatus as claimed in claim 10, including means for processing a new character stroke by including said new stroke in one of said sub-segments when said stroke is within the boundary of said one of said sub-segments and generating a new sub-segment including said new stroke when said new stroke is outside of said boundary.

12. A recognition apparatus as claimed in claims 10 or 11, wherein said merging means and said segment means process at least three previously generated sub-segments

13. A recognition apparatus as claimed in claim 12, wherein said merging means merges said sub-segments using a plurality of predetermined thresholds for said distance.

14. A recognition apparatus as claimed in claim 13, wherein said thresholds represent a range of distance values.

15. A recognition apparatus as claimed in claim 10, including means for merging adjacent segments which produce a merged boundary having a height to width ratio closer to one than the separate boundaries of the adjacent segments.

16. A recognition apparatus as claimed in claim 10, 11, 13, 14 or 15, including character recognition means for performing character recognition on said segments to generate a segment list representing said segments and respective possible characters associated with the segments.

17. A recognition apparatus as claimed in claim 16, including means for generating a time sequence representation of said possible characters from said segment list and for processing said time sequence representation using a language model to generate text representing recognised characters, and means for displaying said text.

18. A recognition apparatus as claimed in claim 11, having a segmentation unit which includes said sub-segment generating means, said merging means, said segment generating means and said new character stroke processing means, and which operates in real-time for each written character stroke.

21. A recognition module as claimed in claims 10 or 20, wherein said merging means and said segment means process at least three previously generated sub-segments.

22. A recognition module as claimed in claim 21, wherein said merging means merges said sub-segments using a plurality of predetermined thresholds for said distance.

23. A recognition module as claimed in claim 22, wherein said thresholds represent a range of distance values.

19. A recognition module stored on a computer readable storage medium, having:

means for generating sub-segments representing at least one character stroke, said sub-segments each having a boundary enclosing the at least one stroke of the sub-segment;

means for merging said sub-segments when the distance between centroids of the sub-segments are less than a predetermined threshold; and

means for generating segments representing possible characters from said sub-segments.

20. A recognition module as claimed in claim 19, including means for processing a new character stroke by including said new stroke in one of said sub-segments when said stroke is within the boundary of said one of said sub-segments and generating a new sub-segment including said new stroke when said new stroke is outside of said boundary.

24. A recognition module as claimed in claim 19, including means for merging adjacent segments which produce a merged boundary having a height to width ratio closer to one than the separate boundaries of the adjacent segments.

25. A recognition module as claimed in claim 19, 20, 22, 23 or 24, including character recognition means for performing character recognition on said segments to generate a segment list representing said segments and respective possible characters associated with the segments.

26. A recognition module as claimed in claim 25, including means for generating a time sequence representation of said possible characters from said segment list and for processing said time sequence representation using a language model to generate text representing recognised characters.

27. A recognition module as claimed in claim 20, having a segmentation module which includes said sub-segment generating means, said merging means, said segment generating means and said new character stroke processing means, and which is executed in real-time for each written character stroke.

28. A computer-readable medium containing executable instructions for performing the steps of:

generating sub-segments representing at least one character stroke, said sub-segments each having a boundary enclosing the at least one stroke of the sub-segment;

merging said sub-segments when the distance between centroids of the sub-segments are less than a predetermined threshold; and

generating segments representing possible characters from said sub-segments

30. A computer-readable medium as claimed in claim 28 wherein the step of generating one of said segments includes processing at least three previously generated sub-segments.

31. A computer-readable medium as claimed in claim 30, wherein said merging step includes merging said sub-segments using a plurality of predetermined thresholds for said distance.

32. A computer-readable medium as claimed in claim 31, wherein said thresholds represent a range of distance values.

33. A computer-readable medium as claimed in claim 28, including the step of merging adjacent segments with produce a merged boundary having a height to width ratio closer to one than the separate boundaries of the adjacent segments.

34. A computer-readable medium as claimed in claims 28, 29, 31, 32 or 35;

including the step of performing character recognition on said segments to generate a segment list representing said segments and respective possible characters associated with said segments.

35. A computer-readable medium as claimed in claim 34, including the step of generating a time sequence representation of said possible characters from said segment list and processing said time sequence representation using a language model to generate text representing recognised characters.

36. A computer-readable medium as claimed in claim 28, wherein said generating and merging steps are executed in the real-time as character strokes are written.