Feature selection based on genetic algorithms and support vector machines for handwritten similar Chinese characters recognition

This paper presents a feature selection approach for handwritten similar Chinese characters recognition. The optimal features can be selected automatically by genetic algorithms from the representations in the form of elastic meshing based on wavelet transform. Three different combinations of binary support vector machines classifiers are discussed when multi-class classification problem must be dealt with. In our approach the fitness scores for different feature subset are derived from the cross-validation rate by using one-against-one strategy based support vector machines classifier with the Gaussian kernel function. The experiment results confirm the effectiveness and practicality of the approach.