REFERENCE DATA OPTIMIZATION LEARNING METHOD AND PATTERN RECOGNITION SYSTEM

摘要

PROBLEM TO BE SOLVED: To automatically learn the recognition of new reference data indispensable to pattern matching processing in a relatively short time.;SOLUTION: The optimal reference data are centroids (center) of all input data where certain reference data become recognition winner ((a) in Figure 3). However, the recognition on line to which input data are serially inputted generally cannot obtain the optimal reference data. Then, the centroids are calculated from input data of a definite number of times corresponding to the reference data regarded as the recognition winner and the reference data are optimized. In order to avoid erroneous recognition and improve the recognition rate, a recognition area is reduced and expanded. By applying optimization like these, even when the input data fluctuate by some influences, changes in distribution can be dealt with sufficiently.;COPYRIGHT: (C)2005,JPO&NCIPI