Image analysis – Pattern recognition – Feature extraction
Patent
1995-08-08
1997-11-18
Boudreau, Leo
Image analysis
Pattern recognition
Feature extraction
382224, G06K 900
Patent
active
056895844
DESCRIPTION:
BRIEF SUMMARY
TECHNICAL FIELD
1. Background of the Invention
The present invention relates to a method of and an apparatus for recognizing input characters and various different input patterns by referencing a prescribed dictionary, and also relates to a method of creating a pattern recognition dictionary used in the same.
2. Background Art
Generally, pattern recognition for recognizing letters, numbers, patterns, etc., is performed by comparing an input pattern with the contents of a dictionary in which standard patterns are stored in advance. Pattern recognition, therefore, has an inherent problem that as the kinds of letters, numbers, patterns, etc., to be recognized increase, the size of the dictionary for storing them becomes large, and the time of the recognition process increases proportionally. Hence, there arises a need to reduce the time required for pattern recognition.
The prior art discloses a variety of pattern recognition methods, including a method based on pattern matching, a method based on feature extraction, etc. In the method based on pattern matching, for example, printed characters, handwritten characters, etc., are optically read by a scanner or the like; each input character is then matched against a plurality of standard patterns stored in a pattern recognition dictionary, and the name of the standard pattern that has the greatest similarity is selected as the name of the input pattern.
In the method based on feature extraction, for example, a dictionary is created in advance in which the vertical and horizontal distributions of character parts, the relations between character elements and their neighbors, etc., are recorded as character features; to identify a captured character, its features are extracted and compared with those of the recorded characters, and the character having the features that provide the greatest similarity is determined as being equivalent to the captured character.
A pattern recognition method using a neurocomputer is also known. In this type of pattern recognition method, there are provided, for example, an input layer, which consists of neurons corresponding to the dots forming the two-dimensional patterns of letters, numerals, various pattern, etc., to be recognized, an output layer, which consists of neurons corresponding to recognition outputs, and an intermediate layer, which provides weighted connections between them; the weighting in the intermediate layer is adjusted using a back propagation method, etc., and, upon completion of learning, the output layer outputs the result of the recognition, such as a pattern name, for the pattern input to the input layer.
The prior art method of recognition using pattern matching requires that preprocessing be performed to normalize the size, position, etc., of an input pattern to those of the standard patterns stored in the dictionary, and also that the input pattern be matched against all the standard patterns stored in the dictionary, and hence, has a shortcoming that an appreciable time is spent in preprocessing and pattern matching.
On the other hand, in the method of recognition using feature extraction, comparison must be made of all the features of the character to be recognized, and the number of features of letters, numerals, and patterns becomes enormous. This presents a problem in that the dictionary size becomes very large if a high recognition rate is to be obtained, and hence there is a shortcoming that the recognition time increases.
Furthermore, in the case of material containing letters, numbers, etc., of different type styles such as gothic and italic, along with standard patterns, pattern matching or extraction and comparison of features needs to be performed for each type style, which means that the same number of matching or comparison operations as the number of type styles must be performed for one character name, requiring a considerable time for recognition. Instead of a method requiring successive comparisons, a method using a technique of broad classification to limit the kinds of patterns to b
REFERENCES:
patent: 5182777 (1993-01-01), Nakayama et al.
patent: 5255342 (1993-10-01), Nitta
patent: 5293429 (1994-03-01), Pizano et al.
patent: 5315668 (1994-05-01), O'Hair
Bird Systems Research Institute, Inc.
Boudreau Leo
Tadayon Bijan
LandOfFree
Method of and apparatus for pattern recognition and method of cr does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of and apparatus for pattern recognition and method of cr, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of and apparatus for pattern recognition and method of cr will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1571942