Image analysis – Pattern recognition – Classification
Reexamination Certificate
1999-03-19
2001-05-15
Tadayon, Bijan (Department: 2621)
Image analysis
Pattern recognition
Classification
C382S310000
Reexamination Certificate
active
06233354
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a character recognition device, and in particular to a character recognition device which inputs an image of a series of characters and outputs the results of recognition of individual characters.
2. Description of the Related Art
When recognizing words given as an image, conventionally, for example, as in Japanese Patent Application, First Publication, No. Hei 6-348911, the word image is converted into two colors, the circumscribing rectangle which circumscribes the black pixel concatenated component is found, these circumscribing rectangles are integrated based on the graphical characteristics such as the distances between the surrounding circumscribing rectangles, the integrated rectangular areas are subject to character recognition, the character recognition result is checked against correct word spellings stored in a dictionary, and as a result the character with the largest degree of agreement is output.
A problem in the above-described conventional technology is that when there is a character with a part missing in the image, it cannot be correctly read. The reason is that when the rectangular area is subject to character recognition, it is assumed that its circumscribing rectangle matches a character frame of a character in the character recognition dictionary. This assumption does not hold when a part of the character is missing, and there is the concern that the result of the character recognition of this character will be obviously lacking validity.
For example, assume that an image including the character ‘Y’ is extracted as shown in FIG.
15
A.
FIG. 15A
is the case that one part of the right side of ‘Y’ is missing. In contrast, assume that the template for the ‘Y’ character type as shown in
FIG. 15C
is provided in the character recognition dictionary.
Generally, the character recognition device calculates the degree of similarity from the match between the circumscribing rectangle in the recognition object image and the character frame of the template in the character recognition dictionary. For example, the degree of similarity is calculated with
FIG. 15C
after
FIG. 15A
is transformed into
FIG. 15B
in order to match the character frame of the template. Because the degree of similarity between
FIG. 15B
after transformation and the template shown in
FIG. 15C
is low, this causes a recognition error.
That is, if an extracted character has a missing part, there is the problem that a correct character recognition result will not be obtained because the similarity with the template of the correct character type is low, or because of the possibility that by chance there may be a high degree of similarity with the template of a character type other than the correct one.
SUMMARY OF THE INVENTION
An object of the present invention is to provide a means for recognizing at a word image having a character with a missing part a higher degree of reliability.
The present invention provides a key character selection means (reference number
22
in
FIG. 1
) which finds the degree of independent confidence based on the recognition result and size information of an extracted character candidate, and selects the character candidate with a high degree of independent confidence as a key character; a doubtful character selection means (reference number
23
in
FIG. 1
) which uses the size information of the key character as a standard, and finds the degree of relative confidence of the relevant character candidate from the results of recognition of other candidate characters and size information, and selects character candidates with a low degree of relative confidence as doubtful characters; and a character reconfirmation means (reference numeral
24
in
FIG. 1
) which uses the size information of the key character as a standard, infers the size information of the selected doubtful character, and carries out extraction and character recognition of the doubtful character once again based on such size information.
The key character selection means selects character candidates to be used as the standard when evaluating the size information of the extracted area of the other character candidates. The doubtful character selection means infers the size information of the extraction area of other character candidates using the size information of the extraction area of the key characters as the standard, and selects candidates having deviating size information as doubtful characters. The character reconfirmation means evaluates the size information of the extraction area of the relevant character candidate with respect to each character candidate selected by the doubtful character selection means, carries out extraction of the relevant character candidate by the input word image based on this size information, and applies character recognition by matching the character frames of the template of the character recognition dictionary with the outer frame of the extracted area.
REFERENCES:
patent: 5164996 (1992-11-01), Pastor
patent: 5504822 (1996-04-01), Holt
patent: 5544260 (1996-08-01), Chefalas et al.
patent: 5883986 (1999-03-01), Kopec et al.
patent: 55-97671 (1980-07-01), None
patent: 6-348911 (1994-12-01), None
Azarian Seyed H.
Foley & Lardner
NEC Corporation
Tadayon Bijan
LandOfFree
Character recognition device does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Character recognition device, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Character recognition device will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2534181