JPS5739478A - Recognition system - Google Patents

Abstract

PURPOSE:To easily extract a significant feature, and to elevate a recognition efficiency of an optical character reader and a voice recognition device, by extracting a significant feature for discrimination in accordance with a selected pseudo-answer. CONSTITUTION:A character pattern from a pattern memory is segmented by a segment detecting circuit 1, adjustment of inclination, etc. are executed by a normalizing circuit 2, and a pattern signal of 1 character portion is outputted. A basic feature of a read pattern of this pattern signal is extracted by a basic feature extracting circuit 8, it is matched with a pseudo-answer of similarity degree dictionary 4 by a similarity degree circuit 3, and a pseudo-data 17 is supplied to a psudo- answer deciding circuit 10. Also, a data 18 of a basic feature extracted by the extracting circuit 8 is inputted to the circuit 10, is collated with a basic feature of a group in a basic feature dictionary 9, is strictly extracted, and a data 19 is supplied to a subfeature extracting circuit 12 and an answer outputting circuit 14. Subsequently, a significant feature is extracted in accordance with a pseudo-answer, by using a subfeature extracting dictionary 11 and a subfeature dictionary 13 by the circuits 12, 14, respectively.