System and method for unconstrained on-line alpha-numerical handwriting recognition

• US 5,841,902 A
• Filed: 10/01/1996
• Issued: 11/24/1998
• Est. Priority Date: 09/27/1994
• Status: Expired due to Term

• Current Assignee: Transpacific IP Group Limited
• Original Assignee: Industrial Technology Research Institute
• Inventors: Tu, Lo-Ting
• Primary Examiner(s): Johns, Andrew W.
• Assistant Examiner(s): Davis, Monica S.
• Application Number:
  US08/725,345
• Time in Patent Office: 784 Days
• Field of Search: 382/187, 382/190, 382/192, 382/193, 382/224, 382/225
• US Class Current: 382/187
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1423 the instrument generating sequences of position coordinates corresponding to handwriting preprocessing or recognising digital ink G06V30/32
• CPC Class Codes: G06V 30/18 Extraction of features or characteristics of the image
• CPC Class Codes: G06V 30/192 using simultaneous comparisons or correlations of the image signals with a plurality of references
• CPC Class Codes: G06V 30/347 Sampling; Contour coding; Stroke extraction
• CPC Class Codes: G06V 30/373 using a special pattern or subpattern alphabet

Abstract

A character recognition system includes a character input device, such as a stylus and tablet or optical scanner, for receiving inputted characters, and a processor. The processor determines which of a number of model characters best matches the inputted character. To that end, the processor compares each inputted character to each of a plurality of classes into which the model characters are organized. Specifically, the processor extracts a feature value vector from the inputted character, and compares it to the mean feature value vector of each class. The processor recognizes the inputted character as the model character corresponding to the mean feature value vector which is closest to the feature value vector of the inputted character. The processor also constructs the database from multiple specimens of each model character. The processor organizes the specimens of each model character into multiple classes. The processor then determines the mean feature value vector of each class.

Claims

1. A method for recognizing a to-be-recognized character using a character recognition database comprising the step of constructing said character recognition database by performing the steps of:

inputting a plurality of character specimens for each model character to be recognized,

for each inputted model character, organizing said specimens into at least one class, wherein the character specimens of at least one model character are organized into plural classes,

extracting values of a feature vector, containing plural features, of each of said inputted character specimens, to produce a feature value vector, containing plural feature values, for each of said character specimens,

for each of said classes, forming a mean feature value vector, containing plural mean feature values, as an unweighted average of said feature value vectors of said character specimens of said respective classes,

storing as said character recognition database said plural mean feature value vectors, including plural mean feature value vectors for said at least one of said model characters for recognizing a to-be-recognized character as a respective model character, and

for at least one model character, adjusting said classes of said character specimens to stabilize said classes for character recognition by reassigning one or more character specimens to different classes of said respective model character and repeating said step of forming said mean feature value vectors of said adjusted classes.

2. The method of claim 1 further comprising recognizing a to-be-recognized character as one of said model characters by performing the steps of:

performing said step of extracting on said to-be-recognized character using said same feature vector,

comparing said feature value vector of said to-be-recognized character to at least some of said mean feature value vectors of said character recognition database, to determine a best matching feature value vector, and

recognizing said to-be-recognized character as the model character corresponding to the class having the best matching mean feature value vector.

3. The method of claim 1 wherein said step of adjusting comprises executing the K_-- means procedure.

4. The method of claim 1 wherein said step of organizing said specimens further comprises the steps of:

comparing the number of said remaining specimens collected into said new class to a second threshold, and

depending on said comparison, adjusting said radius and recollecting, into said new class, each of said remaining specimens within said adjusted radius from said center of said new class.

5. The method of claim 1 wherein said step of determining said center of said new class comprises the step of determining the mean of said feature value vectors of said remaining character specimens.

6. The method of claim 5 wherein said radius determined in said step of determining a radius is a function of the standard deviation of said features of said remaining character specimens.

7. The method of claim 1 further comprising the step of segmenting an image of each of said inputted character specimens into a plurality of zones.

8. The method of claim 7 wherein said step of segmenting produces a plurality of segmented character images for each of said inputted character specimens.

9. The method of claim 7 wherein said step of extracting further the step of:

extracting said feature values from particular zones of particular ones of said segmented character images of each of said character specimens according to a predetermined scheme.

10. The method of claim 7 wherein said image is segmented into variable dimensioned zones.

11. The method of claim 7 wherein said step of determining a feature value vector comprises the step of extracting stroke density function feature values by projecting a plurality of inspection lines across one of said zones of a character specimen, counting the total number of times said inspection lines cross said zone and dividing the total number of times said inspection lines cross said zone by the number of inspection lines projected in said step of projecting.

12. The method of claim 7 wherein said step of determining a feature value vector comprises the step of extracting peripheral background area feature values by evaluating ##EQU11## where:

m is an index of said zones into which said image of said character specimen is segmented,

n is an index of points on an axis relative to an m^th one of said zones which takes on each value from 1 to N,

N is a maximum dimension on said axis,

λ

_n is a distance from an n^th one of said points on said axis to a filled pixel of said m^th zone in pixels,

X_m is a horizontal width of said m^th zone, and

Y_m is a vertical height of the m^th zone.

13. The method of claim 7 wherein said step of determining a feature value vector comprises the step of extracting contour line length features values by evaluating:

##EQU12## where m is an index of said zones into which said image of said character specimen is segmented,

n is an index of points on an axis to an m^th one of said zones, which successively takes on each value from 1 to N,

N is a maximum dimension on said axis,

λ

_n is the distance in pixels from an n^th one of said points on said axis to a filled pixel of said m^th zone in pixels,

X_m is a horizontal width of said m^th zone, and

Y_m is a vertical height of said m^th zone.

14. The method of claim 7 wherein said step of determining a feature value vector comprises the step of extracting gradient feature values by evaluating ##EQU13## where:

m is an index of said zones into which said image of a character specimen is divided,

i is a column index of an m^th one of said zones,

j is a row index of said m^th zone,

Len_ij is a vector of values Len, including one value for each direction code (Dir_ij) generated using equation (7a), for a pixel in said i^th column and j^th row of said m^th zone,

(Dir_ij) is a direction code of a pixel in an i^th column and j^th row of said m^th zone,

Bdd_m (Dir_ij) represents the boundary width of said m-th zone in a direction normal to a direction specified by (Dir_ij), and

X and Y are values generated using the following kernels;

##STR2##

15. The method of claim 2 wherein said step of comparing comprises the step of evaluating a discriminant function:

##EQU14## and where;

q is an index identifying the inputted character,

r and c are independent indexes identifying particular classes and are defined over the range of 1 . . . C

l is an index identifying a particular feature and is defined over the range of 1 . . . L,

L is the total number of features,

C is the total number of classes over the entire database of model characters,

A_ql is an l^th feature value for the q^th input character,

B_rl is an l^th mean feature value for the r^th class,

A_q is the feature value vector of the q^th input character including one feature value (A_q1, A_q2, . . . , A_qL) for each feature 1 to L,

B_r is the mean feature value vector of the r^th class including one mean feature value B_r1, B_r2, . . . , B_m) for each feature 1 to L,

V_lc is the standard deviation of the l^th feature in a given c^th class, and

S_l is the mean standard deviation for the l^th feature over all C classes.

16. The method of claim 15 wherein said q^th inputted character is recognized as a model character corresponding to an r^th class for particular values of q and r which minimize Dis (A_q, B_r).

17. The method of claim 1 wherein said character recognition database includes model phonetic symbols.

18. The method of claim 1 wherein said character recognition database includes alpha-numeric characters.

19. The method of claim 1 wherein said inputted characters are inputted on-line.

20. The method of claim 1 wherein said inputted characters are inputted off-line.

21. The method of claim 19 wherein said inputted characters are handwritten characters.

22. The method of claim 1 wherein said inputted characters are machine printed characters.

23. A system for recognizing a to-be-recognized character using a character recognition database comprising:

a character input device for electronically receiving a plurality of inputted character specimens for each model character to be recognized in constructing a database and

a processor connected to said character input device for electronically constructing a database by electronically organizing said c specimens into a plurality of classes wherein the character specimens of at least one model character are organized into plural classes said processor also for electronically extracting values of a feature vector, containing plural features, of each of said inputted character specimens, to produce a feature value vector, containing plural feature values, for each of said character specimens, for each of said classes, forming a mean feature value vector, containing plural mean feature values, as an unweighted average of said feature value vectors of said character specimens of said respective classes, and storing as said character recognition database said plural mean feature value vectors, including plural mean feature value vectors for said at least one of said model characters for recognizing a to-be-recognized character as a respective model character, and for at least one model character, adjusting said classes of said character specimens to stabilize said classes for character recognition by reassigning one or more character specimens to different classes of said respective model character and repeating said step of forming said mean feature value vectors of said adjusted classes.

24. The character recognition system of claim 23 wherein said character input device electronically receives a to-be-recognized character, and

wherein said processor is also for extracting values of said same feature vector from said to-be-recognized character to produce a feature value vector for said to-be-recognized character, comparing said feature value vector of said to-be-recognized character to at least some of said mean feature value vectors of said character recognition database, to determine a best matching feature value vector, and recognizing said to-be-recognized character as the model character corresponding to the class having the best matching mean feature value vector.

25. The system of claim 23 further comprising a memory for electronically storing said character recognition database.

26. The system of claim 23 wherein said character input device is a stylus and tablet.

27. The system of claim 23 wherein said character input device is an optical scanner.