Printed Character Recognition by Matching Partial Patterns

概要

論文の詳細を見る
A method to recognize printed characters by matching partial patterns is introduced. Those partial patterns like line segments. various corners. or separated segments are extracted from sample patterns and stored in a memory. Values of similarity are calculated by correlating those partial patterns and small regions of input patterns at predetermined positions. The scores are counted on the basis of similarity value and weight value for each character. The input pattern is determined to be the character which gives the max1mum sum of scores. The advantages of this method are that it is not strongly affected by the noises like spots and grains and the required memory capacity is smaller than that needed for the whole pattern matching method. Results of simulation for 40 alphanumeric and symbo1 characters of ISO-B font are shown.
一般社団法人情報処理学会の論文