A Hybrid System for the Recognition of Hand-Written Characters

摘要

In this paper we introduce a new hybrid system for the automated recognition of hand-written characters - we combine the most promising approaches of the last decade, i.e., neural networks and structural/syntactical analysis methods. The given patterns represent handwritten capital letters and digits stored in arrays. The first part of the hybrid system consists of the implementation of a neural network and yields a rapid and reliable pre-selection of the most probable characters the given pattern may represent. Depending on the quality and the special characteristics of the given pattern a flexible set of characters is communicated to the second part of the hybrid system, the structural analysis module. The final decision is based on the evaluation of the presence of features, being characteristic for a specific character, in the underlying pattern. Basically, the structural analysis module consists of graph controlled array grammar systems using prescribed teams of productions. We describe the main parts of the implemented hybrid system and demonstrate the power of our approach.