Classification of Handwritten Numerals Using Modular Neural Networks

摘要

In this paper, we propose a method to classify handwritten numerals using the modular neural networks with image dithering. Basically the 'divide and conquer' strategy is employed. The initial clusters produced by SOM learning are extended to further include the clusters which overlap each other. Each MLP is assigned to each extended cluster. The gating network to combine the decisions of the expert MLP networks is designed and trained on such clusters. To further enhance recognizing capability, the recognition methods using dithering patterns are also advanced. The experimental results demonstrated that the proposed method produces very good recognition performance.