Character recognition processing method and apparatus

• US 8,254,689 B2
• Filed: 01/05/2010
• Issued: 08/28/2012
• Est. Priority Date: 07/25/2007
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Katsuyama, Yutaka, Hotta, Yoshinobu, Fujimoto, Katsuhito
• Primary Examiner(s): Desire, Gregory M
• Application Number:
  US12/652,556
• Publication Number:
  US 20100104192A1
• Time in Patent Office: 966 Days
• Field of Search: 345/441, 382/171, 382/182, 382/190, 382/210, 382/236, 382/305
• US Class Current: 382/190
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06V 30/10 Character recognition

Abstract

This method includes: extracting a feature vector for an input character from a reading result of the input character; calculating distances between the feature vector for the input character and vectors including average vectors stored in a system dictionary storing, for each character, the average vector and distribution information, and feature vectors stored in a user dictionary; extracting the top N character codes in an ascending order of the calculated distances; obtaining second distribution information for the character codes, which are included the user dictionary and in the top N character codes; calculating, for each of the top N character codes, a second distance with the feature vector for the input character, by using, for the character codes, which are included in the user dictionary and in the top N character codes, the second distribution information; and identifying a character code whose second distance is shortest.

Claims

1. A character recognition processing method, comprising:

extracting, by using a computer, a feature vector for an input character from a reading result of said input character;

calculating, by using the computer, distances between said feature vector for said input character and vectors including average vectors stored in a system dictionary storing, for each character, said average vector of said feature vectors and distribution information, and feature vectors stored in a user dictionary storing, for each character, said feature vector;

extracting, by using the computer, a top predetermined number of character codes in an ascending order of the calculated distances;

obtaining, by using the computer, second distribution information for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes;

calculating, by using the computer, for each of said top predetermined number of character codes, a second distance with said feature vector for said input character, by using, for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes, said second distribution information, and by using, for said character codes, which are extracted from said system dictionary and included in said top predetermined number of character codes, the corresponding distribution information in said system dictionary; and

identifying, by using the computer, a character code whose second distance is shortest.

2. The character recognition processing method as set forth in claim 1, wherein said obtaining comprises:

identifying a character code in said system dictionary, which has said average vector whose distance with said feature vector for said input character is shortest; and

extracting, as said second distribution information, distribution information corresponding to the identified character code from said system dictionary.

3. The character recognition processing method as set forth in claim 1, wherein said obtaining comprises:

calculating reference distances between said feature vector of each of said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes, and said average vector of said feature vectors of each said character code in said system dictionary;

second identifying a character code in said system dictionary, whose reference distance is shortest; and

reading out, as said second distribution information, distribution information corresponding to said character code identified at said second identifying and is stored in said system dictionary.

4. The character recognition processing method as set forth in claim 3, wherein said predetermined reference distribution information is distribution information for characters used when creating said system dictionary.

5. The character recognition processing method as set forth in claim 1, wherein said second distribution information is predetermined reference distribution information.

6. The character recognition processing method, as set forth in claim 1, wherein said obtaining comprises:

generating, as said second distribution information, synthesized distribution information for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes, from said distribution information, which is stored in said system dictionary and identified based on at least one of said feature vectors corresponding to said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes, and said feature vector for said input character.

7. The character recognition processing method as set forth in claim 6, wherein said generating comprises:

identifying a character code, which is included in said system dictionary and has an average vector whose distance with said feature vector for said input character is shortest, to extract, as third distribution information, distribution information corresponding to the identified character code, from said system dictionary;

calculating reference distances between said feature vector of said character code, which is extracted from said user dictionary and included in said top predetermined number of character codes, and said average vectors of said feature vectors of the respective character codes in said system dictionary to extract, as fourth distribution information, distribution information of said character code, which is extracted from said system dictionary, and whose reference distance is shortest; and

synthesizing said third distribution information and said fourth distribution information to generate said second distribution information.

8. The character recognition processing method as set forth in claim 6, wherein said generating comprises:

identifying a character code, which is included in said system dictionary and has an average vector whose distance with said feature vector for said input character is shortest, to extract, as fifth distribution information, distribution information corresponding to the identified character code from said system dictionary; and

synthesizing said fifth distribution information and predetermined distribution information to generate said second distribution information.

9. The character recognition processing method as set forth in claim 6, wherein said generating comprises:

calculating second reference distances between said feature vector of said character code, which is extracted from said user dictionary and included in said top predetermined number of character codes, and said average vectors of said feature vectors of the respective character codes in said system dictionary to extract, as seventh distribution information, distribution information of said character code, which is included in said system dictionary, and whose second reference distance is shortest; and

synthesizing said seventh distribution information and predetermined reference distribution information.

10. A computer-readable storage medium storing a program for causing a computer to execute a character recognition process, comprising:

extracting a feature vector for an input character from a reading result of said input character;

calculating distances between said feature vector for said input character and vectors including average vectors stored in a system dictionary storing, for each character, said average vector of said feature vectors and distribution information, and feature vectors stored in a user dictionary storing, for each character, said feature vector;

extracting a top predetermined number of character codes in an ascending order of the calculated distances;

obtaining second distribution information for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes;

calculating, for each of said top predetermined number of character codes, a second distance with said feature vector for said input character, by using, for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes, said second distribution information, and by using, for said character codes, which are extracted from said system dictionary and included in said top predetermined number of character codes, the corresponding distribution information in said system dictionary; and

identifying a character code whose second distance is shortest.

11. A character recognition processing apparatus, comprising:

a first extracting unit to extract a feature vector for an input character from a reading result of said input character;

a first calculating unit to calculate distances between said feature vector for said input character and vectors including average vectors stored in a system dictionary storing, for each character, said average vector of said feature vectors and distribution information, and feature vectors stored in a user dictionary storing, for each character, said feature vector;

a second extracting unit to extract a top predetermined number of character codes in an ascending order of the calculated distances;

an obtaining unit to obtain second distribution information for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes;

a second calculating unit to calculate, for each of said top predetermined number of character codes, a second distance with said feature vector for said input character, by using, for said character codes, which are extracted from said user dictionary and included in said top predetermined number of character codes, said second distribution information, and by using, for said character codes, which are extracted from said system dictionary and included in said top predetermined number of character codes, the corresponding distribution information in said system dictionary; and

an identifying unit to identify a character code whose second distance is shortest.