(The) recognition of korean characters by the structural approach

Abstract

This thesis proposes the structural method of Korean character recognition and representation. Simple convex polygons are selected as primitives and their positional relationships are represented as labeled digraphs. Segmentation, labeling regions, boundary tracing, corner detection, and hole finding processes are performed as preprocessings for the primitive extraction. For concavity finding and decomposition, the algorithms based on the computational geometry are developed. Each character is represented as digraphs called Primitive Connection Digraph (PCD) by the positional relationship and connection relationship between the primitive convex polygons. Each PCD is represented as triples called Positional Relation Triples(PRT). The positional relationships are obtained by calculating the center of area and other shape informations. The classification step can be performed by the syntax analysis (parsing) or decision procedure. In this thesis, the rule-based decision procedure method was used. As it consists of IF-THEN-ELSE rule structures, it can be easily modified and be easily proliferated.