Image analysis – Pattern recognition – Context analysis or word recognition
Reexamination Certificate
1997-03-21
2001-02-06
Bella, Matthew C. (Department: 2721)
Image analysis
Pattern recognition
Context analysis or word recognition
C382S105000, C382S177000
Reexamination Certificate
active
06185338
ABSTRACT:
BACKGROUND OF THE INVENTION
I. Field of the Invention
The present invention concerns character recognition.
II. Related Art and Other Considerations
Japanese Laid-Open Patent Publication No. 2-206894 proposes a character recognition method that recognizes characters of, e.g., a license plate of an automobile by binarizing an image of character areas on the license plate and seeking a 4-digit serial number composed of relatively large characters included in the binary image by key features (e.g., width of characters and gap between characters) of the license plate.
Japanese Laid-Open Patent Publication No. 6-131492 proposes a character recognition method that searches an area by relative positions determined from information on position of a 4-digit serial number and then recognizes characters in each of the determined character areas.
In the former method it is difficult to accurately determine the position of a serial number if any positional information on any digit of the 4-digit serial number could not correctly be obtained due to (1) noise caused by an incorrect positioning or conditioning of an image-taking camera or (2) the obtained serial number containing only 1 digit or 2 digits, i.e., a reduced number of large-size characters.
The latter method also requires the accurate determination of the serial number because its location is a key to determine areas of the other character areas. For example, in this latter method it is difficult to correctly presume character areas in an image of a license plate if the input image lacks a part of a license plate or does not indicate a part of a serial number due to an obstruction or it includes noise signals caused by both side frames, an indicator lamp and so on.
SUMMARY OF THE INVENTION
The present invention relates to a character recognizing method which is used for extracting characters (including numerals) indicated for example on a license plate of an automobile and recognizing the extracted characters.
In view of the foregoing drawbacks of the conventional character recognizing method which extracts serial number and presumes other character areas on the basis of relative positions of the extracted serial number area, the present invention is made to provide a character recognition method which is capable of accurately recognizing all characters contained in an object image of, e.g., a license plate by means of previously setting a certain number of area-frame candidates according to known arrangement of serial-number digit areas, a use-code area, a land-transport administration branch office-name/classification number area, selecting one of the area-frame candidates, which has the minimum inconformity in relative positions of characters in respective character areas, determining respective character areas on the selected area-frame and then recognizing each character in each character area.
An object of the present invention is to provide a character recognizing method which comprises the steps of: extracting information on each of rectangles circumscribing respective characters contained in an inputted binary quantized image; selecting least disagreed one of candidates of area frames preset according to a known format of character areas on a recognition object by comparing each of the candidates with the extracted character-circumscribed rectangles as to inconformity (contradiction) in relative arrangement of the rectangles to character areas on an area frame; and recognizing characters indicated in the s elected character areas respectively. A character recognition object (e.g., a license plate of an automobile) has a certain known format of character areas. However, the same character-recognition objects may have a variety of sizes and relative positions of character areas as well as a plurality of area-frame candidates. As this method compares each of area-frame candidates with extracted information on character-circumscribing rectangles and selects such an area-frame candidate that has the least inconformity of positions of their character areas relative to the rectangles, the selected character areas may have a correct relation with the extracted rectangles even if an input image does not include a part of the character recognition object or lacks a part of information due to an obstruction or it contains noise signals from peripheral portion of the object. Thus, the method can attain a high accuracy of character recognition since it recognizes each of characters on the character area candidate selected according to the least inconformity criterion.
Another object of the present invention is to provide a character recognition method which includes the content of the invention above mentioned and is further featured in that the step of selecting least disagreed one of area-frame candidates consists of calculating a degree of inconformity in properties and conditions of character areas of candidate to the extracted information on character-circumscribed rectangles, determining a total of the calculated inconformity values of the character areas and selecting an area frame having the least inconformity. This method can attain a considerably improved accuracy of selecting a correct area frame.
A further object of the present invention is to provide a character recognition method which includes the content of the invention first-mentioned and is further featured in that the step of recognizing a string of characters selects a character string having the largest average value of similarities of compare d characters. A character string consists of a known and considerable limited combination of characters. Therefore, this method can attain an increased accuracy of recognizing a character string owing to selecting a character string whose characters have the largest value of similarity with those of the previously determined character string.
REFERENCES:
patent: 4567609 (1986-01-01), Metcalf
patent: 4802231 (1989-01-01), Davis
patent: 5065440 (1991-11-01), Yoshida et al.
patent: 5081685 (1992-01-01), Jones, III et al.
patent: 5285505 (1994-02-01), Kim et al.
patent: 5315664 (1994-05-01), Kumagai
patent: 5425108 (1995-06-01), Hwang et al.
patent: 5502777 (1996-03-01), Ikemure
patent: 5600736 (1997-02-01), Hagita
patent: 5651075 (1997-07-01), Frazier et al.
patent: 5734750 (1998-03-01), Arai et al.
patent: 5754685 (1998-05-01), Takahashi et al.
patent: 5784501 (1998-07-01), Tanaka
patent: 41 35 881 A1 (1993-05-01), None
patent: 0 045 549 A1 (1982-02-01), None
patent: 2-206894 (1990-08-01), None
patent: 6-131492 (1994-05-01), None
patent: 92/06449 (1992-04-01), None
Bella Matthew C.
Nixon & Vanderhye P.C.
Sharp Kabushiki Kaisha
LandOfFree
Character recognition using candidate frames to determine... 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 using candidate frames to determine..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Character recognition using candidate frames to determine... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2573094