System and method for segmentation and recognition of patterns

• US 4,654,873 A
• Filed: 10/30/1985
• Issued: 03/31/1987
• Est. Priority Date: 11/27/1982
• Status: Expired due to Term

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Kadota, Shozo, Kurino, Kiyomichi, Nakano, Yasuaki, Komatsu, Hitoshi, Fujisawa, Hiromichi
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Mancuso, Joseph
• Application Number:
  US06/792,987
• Time in Patent Office: 517 Days
• Field of Search: 382/9, 382/30, 382/37, 382/38, 382/48, 382/60
• US Class Current: 382/178
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/153 using recognition of characters or words

Abstract

Pattern segmentation and recognition in which hand-written characters are transformed electrically into 2-dimensional image patterns, wherein if ambiguity exists in segmenting a unit pattern including a character from the image patterns, character recognition is not made compulsively, but a plurality of possible unit patterns are first established. Then, the various unit patterns are segmented, and each unit pattern is identified to be a partial pattern, linked patterns, etc., so that each character is recognized on a basis of total judgement, whereby ambiguity of segmentation is resolved.

Claims

1. A method of segmentation and recognition of patterns comprising the steps of:

(a) transforming handwritten characters into electrical image patterns;

(b) extracting individual pattern elements from said image patterns on the basis of the continuity of the data in said image patterns;

(c) determining the attributes of each of said individual pattern elements;

(d) forming one or more unit patterns each consisting of one or more individual pattern elements based on the attributes of said pattern elements and the positional relationship of said pattern elements in said handwritten characters;

(e) comparing each of said unit patterns with reference patterns that include partial patterns and combination patterns, wherein a partial pattern is less than a complete character and a combination pattern is more than a single complete character;

(f) generating result codes, including the character name of a reference pattern, for each comparing of a unit pattern with said reference patterns; and

(g) recognizing said handwritten characters in accordance with said result codes.

2. A method of pattern segmentation and recognition according to claim 1, wherein said reference patterns include complete patterns, partial patterns, partial patterns in contact with other complete patterns, and patterns in contact with each other.

3. A method of pattern segmentation and recognition according to claim 2, wherein said comparing step comprises the steps of:

producing signals indicating incompleteness and the character names of recognized patterns if said unit pattern is determined to be a part of pattern representing one character;

producing signals indicative of partial contact and the character names of recognized patterns if said unit pattern is determined to be a partial pattern in contact with another complete pattern;

producing signals indicative of contact and the character names of recognized patterns if said unit pattern is determined to be a plurality of patterns in contact with each other; and

producing signals indicative of completeness and the character name of recognized patterns if said unit pattern is determined to be complete as a pattern.

4. A method of pattern segmentation and recognition according to claim 3, wherein said step of recognizing said handwritten characters includes applying rewriting rules to said plurality of result codes to obtain a final result.

5. A method of pattern segmentation and recognition according to claim 1, wherein said comparing step comprises the steps of:

producing signals indicating incompleteness and the character names of recognized patterns if said unit pattern is determined to be a part of a pattern representing one character;

producing signals indicative of partial contact and the character names of recognized patterns if said unit patern is determined to be a partial pattern in contact with another complete pattern;

producing signals indicative of contact and the character names of recognized patterns in said unit pattern is determined to be a plurality of patterns in contact with each other; and

producing signals indicative of the completeness and the character name of a recognized pattern if said unit pattern is determined to be complete as a pattern.

6. A method of pattern segmentation and recognition according to claim 5, wherein said step of recognizing said handwritten characters includes applying rewriting rules to said plurality of unit patterns into result codes to obtain a final result.

7. A system of segmentation and recognition of patterns comprising:

(a) first means for transforming handwritten characters into electrical image patterns;

(b) second means for extracting individual pattern elements from said image patterns on the basis of the continuity of the data in said image patterns;

(c) third means for calculating the attributes of each of said individual pattern elements;

(d) a form dictionary storing information concerning the positional relationship of said pattern elements in said handwritten characters;

(e) fourth means for forming unit patterns from said individual pattern elements based on the attribute information obtained from said third means and information obtained from said form dictionary;

(f) a pattern dictionary storing pattern information, wherein said information includes partial patterns and combination patterns, a partial pattern being less than a complete character, and a combination pattern being more than a single complete character;

(g) fifth means for matching each of said unit patterns with reference patterns obtained from said pattern dictionary and for generating result codes including the character name of a reference pattern;

(h) sixth means for recognizing said handwritten characters in accordance with said result codes from said fifth means.

8. A system of segmentation and recognition according to claim 7, wherein said pattern dictionary comprises means for storing information identifying complete patterns, partial patterns, partial patterns in contact with other complete patterns, and patterns in contact with each other.

9. A system of pattern segmentation and recognition according to claim 8 wherein said fifth means comprises means for:

producing signals indicating incompleteness and the character names of recognized patterns if said unit pattern is determined to be a part of pattern representing one category;

producing signals indicative of partial contact and the character names of recognized patterns if said unit pattern is determined to be a partial pattern in contact with another complete pattern;

producing signals indicative of contact and the character names of recognized patterns if said unit pattern is determined to be a plurality of patterns in contact with each other; and

producing signals indicative of completeness and the character name of a recognized pattern if said unit pattern is determined to be complete as a pattern.

10. A system of pattern segmentation and recognition according to claim 9, further comprising means for transforming said plurality of unit patterns into a one or more of recognition result codes, a final result being obtained by application of rewriting rules to said recognition result.

11. A system of segmentation and recognition according to claim 7, wherein said fifth means comprises means for:

producing signals indicating incompleteness and the character names of recognized patterns if said unit pattern is determined to be part of a pattern representing one character;

producing signals indicative of partial contact and the character names of recognized patterns if said unit pattern is determined to be a partial pattern in contact with another complete pattern;

producing signals indicative of contact and the character names of recognized patterns if said unit pattern is determined to be a plurality of patterns in contact with each other; and

producing signals indicative of completeness and the character name of a recognized pattern if said unit pattern is determined to be complete as a pattern.

12. A system of pattern segmentation and recognition according to claim 11, further comprising means for transforming said plurality of unit patterns into at least one recognition result code, a final result being obtained by application of rewriting rules to said recognition result.

13. A system of pattern segmentation and recognition according to claim 7, wherein said sixth means includes a rewriting rule dictionary for transforming a plurality of result codes into a character result signal.

14. A method of segmentation and recognition of patterns comprising the steps of:

(a) transforming written characters into electrical image patterns;

(b) extracting individual pattern elements including elements representing less than a complete character from said image patterns on the basis of the continuity of the data in said image patterns;

(c) determining the attributes of each of said individual pattern elements;

(d) forming a plurality of unit patterns representing different combinations of the pattern elements based on the attributes of said pattern elements and the positional relationship of said pattern elements of adjacent written characters;

(e) comparing each of said unit patterns with reference patterns including partial patterns, each of the partial patterns being less than a complete character;

(f) generating result codes, including the character name of a reference pattern, for each comparing of a unit pattern with said reference patterns; and

(g) recognizing said written characters in accordance with said result codes.

15. A method of recognition of electrical image patterns representing character patterns comprising the steps of:

(a) extracting individual pattern elements including elements representing less than a complete character from said electrical image patterns on the basis of the continuity of said image patterns;

(b) determining the attributes of each of said individual pattern elements;

(c) forming a plurality of character lists of unit patterns based on the attributes of said pattern elements and the positional relationship of said pattern elements in which the pattern elements are grouped hypothetically into one or more unit patterns, each unit pattern consisting of one or more individual pattern elements;

(d) checking said character lists as to whether each of said unit patterns satisfies any of predetermined conditions relating to the character patterns;

(e) selecting one of said character lists based on the result of said checking.