Image analysis – Pattern recognition – On-line recognition of handwritten characters
Reexamination Certificate
1996-09-17
2003-12-30
Au, Amelia M. (Department: 2623)
Image analysis
Pattern recognition
On-line recognition of handwritten characters
C382S227000
Reexamination Certificate
active
06671403
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to a pattern recognition method and apparatus, and a computer controller for converting an input unknown pattern into a code or the like by comparing the input pattern with a standard pattern prepared in advance and, more particularly, to a pattern recognition method and apparatus, and a computer controller for acquiring a recognition candidate using a plurality of recognition methods with respect to an input unknown pattern.
In general, a large number of techniques for converting an input unknown pattern into a code by comparing the input pattern with a standard pattern pre-stored in a recognition dictionary or the like have been proposed. For example, in the field of on-line handwritten recognition, a vector matching method in which a set of coordinate points of handwritten strokes input from, e.g., a tablet is processed as an unknown pattern, the strokes are converted into vectors, and the vectors are compared with basic vectors in a recognition dictionary to derive a recognition result, a feature point matching method in which strokes constituting a character are approximated by some feature points (representative points), and the distances between the approximated feature points and those of a standard pattern in a recognition dictionary are calculated to derive a recognition result, and the like are representative methods. In such recognition methods, the order of candidates of character codes obtained by recognition processing is determined using the similarity (a recognition calculation value or the like) between an input character pattern and a standard pattern in a recognition dictionary.
Recently, the following method has been proposed. This method uses a combination of a basic recognition process using a standard recognition dictionary pre-stored in a recognition apparatus, and a registered recognition process for improving the recognition rate by additionally registering or changing character data unique to a user and corresponding character codes as a personal dictionary, thereby outputting a recognition result.
However, in this prior art, when the basic recognition process and the registered recognition process are used in combination, if these two processes have different algorithms, a problem is posed upon outputting candidates. For example, assume that A represents a recognition result of an input unknown pattern in the basic recognition process, and B represents a recognition result in the registered recognition process. In this case, a candidate sequence must be generated by combining A and B as a final recognition result. However, since the recognition results A and B are output results of different algorithms, and have different levels of recognition calculation values, a candidate order sequence cannot be normally generated by merely comparing the calculation values.
SUMMARY OF THE INVENTION
The present invention has been made in consideration of the above-mentioned problem and has as its object to provide a pattern recognition method and apparatus, which can arrange recognition results obtained by individual recognition methods in an appropriate order when recognition results obtained by applying a plurality of different recognition methods to a certain unknown pattern are to be combined and output.
It is another object of the present invention to improve the recognition precision by appropriately determining the recognition candidate order.
In order to achieve the above objects, a pattern recognition apparatus according to the present invention comprises the following arrangement.
That is, there is provided a pattern recognition apparatus for performing pattern recognition by calculating a similarity amount between a pattern to be recognized and each of standard patterns which are registered in advance, comprising recognition means for performing a plurality of different recognition processing operations for the pattern to be recognized, and acquiring one or a plurality of recognition candidates and similarity amounts in each recognition processing operation, conversion means for converting the similarity amounts obtained by the respective recognition processing operations of the recognition means into similarity amounts based on a scale common to the plurality of different recognition processing operations, and generation means for generating a recognition candidate sequence by determining an order of the recognition candidates obtained by the plurality of different recognition processing operations on the basis of the similarity amounts converted by the conversion means.
Preferably, the apparatus further comprises storage means for storing a function representing a relationship between similarities and corresponding accuracies for each of the plurality of different recognition processing operations, and the conversion means converts the similarity amounts obtained by each recognition processing operation of the recognition means into accuracies using the function stored in the storage means, and uses the converted accuracies as the similarity amounts in the common scale. The similarity amounts expressed using scales inherent to the respective recognition processing operations can be easily converted into accuracies, and the accuracies of recognition candidates obtained from the plurality of recognition processing operations can be normally grasped.
Preferably, the function stored in the storage means is a linear function. Since the relationship between the similarity and the accuracy is approximated by a linear function, the processing contents can be simplified.
Preferably, the apparatus further comprises storage means for storing a table for associating states of N-th and (N+1)-th similarity amounts to an accuracy for each of the plurality of different recognition processing operations, and the conversion means acquires an N-th accuracy on the basis of the N-th and (N+1)-th similarity amounts obtained by each recognition processing operation of the recognition means with reference to the table stored in the storage means, and uses the acquired accuracy as the similarity amount in the common scale. In consideration of the difference between N-th and (N+1)-th similarity amounts, the recognition states can be reflected in the accuracies, and conversion into accuracies can be performed more appropriately.
Preferably, the table stored in the storage means registers a relationship between states of first and second similarity amounts and an accuracy corresponding to the first similarity amount, and the conversion means acquires the N-th accuracy by substituting the N-th and (N+1)-th similarity amounts in the first and second similarity amounts in the table to obtain an accuracy, and multiplying the obtained accuracy with an (N−1)-th accuracy. The memory capacity can be saved since all the similarity amounts can be converted into accuracies by storing only a table that stores a correspondence between the states of the first and second similarity amounts and accuracies for each processing.
REFERENCES:
patent: 3764980 (1973-10-01), Dansac et al.
patent: 4484348 (1984-11-01), Shizuno
patent: 4607386 (1986-08-01), Morita et al.
patent: 4760604 (1988-07-01), Cooper et al.
patent: 5048113 (1991-09-01), Yamagata et al.
patent: 5265174 (1993-11-01), Nakatsuka
patent: 5392367 (1995-02-01), Hsu et al.
patent: 5410612 (1995-04-01), Arai et al.
patent: 5481625 (1996-01-01), Suzuki
patent: 5519786 (1996-05-01), Courtney et al.
patent: 5621809 (1997-04-01), Bellegarda et al.
patent: 5960393 (1999-09-01), Cohrs et al.
Arai Tsunekazu
Takasu Eiji
Yoshii Hiroto
Au Amelia M.
Miller Martin
LandOfFree
Pattern recognition apparatus and method utilizing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Pattern recognition apparatus and method utilizing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Pattern recognition apparatus and method utilizing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3103769