Pattern fusion in feature recognition neural networks forhandwritten character recognition

摘要

B. Hussain and M.R. Kabuka (1994) proposed a feature recognitionnneural network to reduce the network size of neocognitron. However, andistinct subnet is created for every training pattern. Therefore, a bignnetwork is obtained when the number of training patterns is large.nFurthermore, recognition rate can be hurt due to the failure ofncombining features from similar training patterns. We propose annimprovement by incorporating the idea of fuzzy ARTMAP in the featurenrecognition neural network. Training patterns are allowed to be merged,nbased on the measure of similarity among features, resulting in a subnetnbeing shared by similar patterns. Because of the fusion of trainingnpatterns, network size is reduced and recognition rate isnincreased