CHARACTER RECOGNITION APPARATUS AND METHOD BASED ON CHARACTER ORIENTATION

• US 20110211759A1
• Filed: 02/28/2011
• Published: 09/01/2011
• Est. Priority Date: 02/26/2010
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: PARK, Jeong-Wan, Kim, Do-Hyeon, Oh, Sang-Wook, Ahn, Hee-Bum
• Granted Patent: US 8,761,514 B2
• Time in Patent Office: Days
• US Class Current: 382/182
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/182 by coding the contour of the pattern
• CPC Class Codes: G06V 30/293 of characters other than Kanji, Hiragana or Katakana

Abstract

A character recognition apparatus and method based on a character orientation are provided, in which an input image is binarized, at least one character area is extracted from the binarized image, a slope value of the extracted at least one character area is calculated, the calculated slope value is set as a character feature value, and a character is recognized by using a neural network for recognizing a plurality of characters by receiving the set character feature value. Accordingly, the probability of wrongly recognizing a similar character decreases, and a recognition ratio of each character increases.

Claims

1. A character recognition apparatus, comprising:

a binarizer for binarizing an input image;

a character extractor for extracting at least one character area from the binarized image;

a character feature value extractor for calculating a slope value of the extracted at least one character area and setting the calculated slope value as a character feature value; and

a character recognizer for recognizing a character by using a neural network for recognizing a plurality of characters by receiving the set character feature value.

2. The character recognition apparatus of claim 1, wherein the character feature value extractor divides the at least one character area into a plurality of block areas, calculates character slope direction information for each of the plurality of divided block areas, gathers the character slope direction information calculated for each of the block areas, and sets the gathered value as a character feature value for each character area.

3. The character recognition apparatus of claim 2, wherein the character feature value extractor presets representative values respectively corresponding to a plurality of slope ranges, calculates a character slope value for each of the plurality of divided block areas, counts the number of selection times of a preset slope range corresponding to a slope range in which the calculated character slope values is included from among the plurality of slope ranges, normalizes a count value accumulated for each of the plurality of slope ranges after completing the count for the plurality of divided block areas, and sets the normalized values as a character feature value.

4. The character recognition apparatus of claim 3, wherein the character feature value extractor calculates the character slope value by using a horizontal Kirsch mask and a vertical Kirsch mask.

5. The character recognition apparatus of claim 1, wherein the neural network for recognizing a plurality of characters comprises:

a character group classification neural network for classifying a character group according to the received character feature value; and

character recognition neural networks for recognizing a character according to the received character feature value.

6. The character recognition apparatus of claim 5, wherein the character recognizer outputs a character group matching score for a character group including a character having the received character feature value if the received character feature value is input to the character group classification neural network, and outputs a character matching score for the having the received character feature value, if the received character feature value is input to the character recognition neural networks.

7. The character recognition apparatus of claim 6, wherein the character recognizer outputs a plurality of character group matching scores through the character group classification neural network, compares the greatest of the plurality of output character group matching scores with a preset threshold, if the greatest character group matching score is greater than or equal to the preset threshold, delivers the character feature value to a corresponding character recognition neural network, and determines the greatest of character matching scores output from the corresponding character recognition neural network as a final result value for the character recognition.

8. The character recognition apparatus of claim 1, wherein, if the greatest character group matching score is less than the preset threshold, the character recognizer compares the character matching score output after inputting the character feature value to the corresponding character recognition neural network with a character matching score corresponding to the second greatest character group, determines whether the greatest character matching score is greater than or equal to a value obtained by subtracting a predetermined value from the character matching score corresponding to the second greatest character group, and if the greatest character matching score is greater than or equal to the value obtained by subtracting the predetermined value from the character matching score corresponding to the second greatest character group, outputs the greatest character matching score as a final result value for the character recognition.

9. The character recognition apparatus of claim 8, wherein, if the greatest character matching score is less than the value obtained by subtracting the predetermined value from the character matching score corresponding to the second greatest character group, the character recognizer outputs the character matching score corresponding to the second greatest character group as a final result value for the character recognition.

10. A character recognition method comprising:

binarizing an input image;

extracting at least one character area from the binarized image;

calculating a slope value of the extracted at least one character area and setting the calculated slope value as a character feature value; and

recognizing a character by using a neural network for recognizing a plurality of characters by receiving the set character feature value.

11. The character recognition method of claim 10, wherein setting the character feature value comprises:

dividing the at least one character area into a plurality of block areas; and

calculating character slope direction information for each of the plurality of divided block areas, gathering the character slope direction information calculated for each of the block areas, and setting the gathered value as a character feature value for each character area.

12. The character recognition method of claim 11, wherein setting the character feature value comprises:

presetting representative values respectively corresponding to a plurality of slope ranges;

calculating a character slope value for each of the plurality of divided block areas;

counting the number of selection times of a preset slope range corresponding to a slope range in which the calculated character slope values is included from among the plurality of slope ranges; and

normalizing a count value accumulated for each of the plurality of slope ranges after completing the count for the plurality of divided block areas and setting the normalized values as a character feature value.

13. The character recognition method of claim 12, wherein calculating the character slope value comprises:

calculating the character slope value by using a horizontal Kirsch mask and a vertical Kirsch mask.

14. The character recognition method of claim 10, wherein the neural network comprises:

a character group classification neural network for classifying a character group according to the received character feature value; and

character recognition neural networks for recognizing a character according to the received character feature value.

15. The character recognition method of claim 14, wherein recognizing the character comprises:

outputting a plurality of character group matching scores through the character group classification neural network;

comparing the greatest of the plurality of output character group matching scores with a preset threshold;

if the greatest character group matching score is greater than or equal to the preset threshold, delivering the character feature value to a corresponding character recognition neural network and determining the greatest of character matching scores output from the corresponding character recognition neural network as a final result value for the character recognition.

16. The character recognition method of claim 15, further comprising:

if the greatest character group matching score is less than the preset threshold, comparing a plurality of character matching scores output after inputting the character feature value to the corresponding character recognition neural network with a character matching score corresponding to the second greatest character group;

if the greatest of the plurality of character matching scores is greater than or equal to a value obtained by subtracting a predetermined value from the character matching score corresponding to the second greatest character group, outputting the greatest character matching score as a final result value for the character recognition.

17. The character recognition method of claim 16, further comprising:

if the greatest character matching score is less than the value obtained by subtracting the predetermined value from the character matching score corresponding to the second greatest character group, outputting the character matching score corresponding to the second greatest character group as a final result value for the character recognition.