Study on an approach of feature selection for similar handwritten Chinese characters recognition

A feature selection approach for similar handwritten Chinese characters recognition is presented in this paper, which is based on genetic algorithms and support vector machines classifier. The technique of combining wavelet transform with elastic meshing is employed for a given handwritten character to extract its feature. The optimal features are selected by genetic algorithms and the problem of determining partial space of similar characters automatically is solved. The experiment results confirm the conclusion that the generalization performance of cross-validation fitness measure is better than that of in-sample validation one.