Parallel algorithms for recognizing handwritten characters using shape features

摘要

Systolic algorithms suitable for VLSI implementation for recognizing handwritten characters using shape features are presented. Local shape features, namely start and end points, edge types and their join-relations in the contours of a given character, are first extracted using a systolic algorithm. The global features consisting of the actual sequence of feature points in the contours are then constructed from the local features using a novel systolic ranking algorithm designed for randomized linked lists. By training the algorithms on a set of 1200 handwritten numerals (120 per digit) a classification scheme is developed. A simple PLA like classifier is also presented. Handwritten numerals are recognized using a horizontal, a vertical and a diagonal scan. The scheme works well even if the images are rotated by an angle between −15 and +15 degrees.