The objective of this paper is to propose a modified neocognitron with production rules embedded for handwritten character recognition. Structured information about the basic features in a character is stored in the production rules constructed by users. A mapping scheme is used to map these rules into the connection weights of the neocognitron. The ability to represent structured information for characters using production rules provides some insights into how this structured information or knowledge can be processed by the network for its character recognition or refinement in the case where a character is misrecognized. The whole process can be controlled by users by analyzing the results of the recognition by refining the production rules to improve the recognition rate. It is much more flexible, and can be used as tools to build a rapid prototype of a pattern recognizer with fault diagnosis capability.
展开▼
