PATTERN RECOGNITION AND LIMITED CHARACTER STRING RECOGNITION SYSTEM USING NEURO, AND CHARACTER PATTERN SORTING DEVICE

摘要

PROBLEM TO BE SOLVED: To effectively recognize and sort characters with high accuracy by extracting the feature of a symbol to be sorted via a 1st neuro computer (neuro) and discriminating a category by a 2nd neuro based on the extracted feature of the said symbol. SOLUTION: The 1st neuro computer 20 having plural feature extraction neuros inputs the slab value in parallel to each other from a preprocessing part 10 and outputs the existence degree of each feature based on the neuro weight, which is included in a symbol to be sorted and previously learnt for every feature. A 2nd neuro computer 30 directly inputs the output of the neuro computer 20 as the new slab value and outputs the separate arithmetic value for every category based on the neuro weight that is previously learnt in every category. Then, a judging part 40 judges the category of an input symbol based on the separate arithmetic value that is outputted from the neuro computer 30 for every category. Thus, even a deformed character group including the handwritten characters and a group of characters having the position shifts, rotations, different sizes, etc., can be effectively discriminated and sorted with high accuracy.