WORD RECOGNITION SYSTEM BY ARTIFICIAL NEURAL NETWORK

摘要

PURPOSE:To prevent a network from being enormous, to improve classifying ability by fine adjustment and the correction of an error and to obtain accurate word information from pattern information by distributing an artificial neural network to networks having different function and constituting the networks in a hierarchical state. CONSTITUTION:Self-organization by learning is achieved in pattern characteristic extraction networks 1a, 1b and 1c, so that pattern characteristic is extracted from the pattern information. A word information extraction network 2 is provided with the self-organization by learning and a function which can be artificially corrected, and extracts the word information from the pattern characteristic. An association network 3 corrects an error in the extracted word information by association before and after the word to extract the accurate word information. By distributing the network to the networks having the different purpose and function and constituting them in the hierarchical state, a learning time is shortened and the network is prevented from being enormous. Thus, the error in the word information is corrected and the classifying ability is improved.