Recognition of ancient Chinese characters based on hybrid kernel WLS-SVR

The shapes of ancient Chinese characters are often uncertain, which reduces the accuracy of recognition by many classifiers. To solve this problem, a new recognition algorithm combining adaptive weighted least squares support vector regression(WLS-SVR) with hybrid kernel function was proposed to recognize ancient Chinese characters. The weight coefficients of WLS-SVR decayed at a rate of the exponential function of prediction errors. The hybrid kernel was constructed using the wavelet kernel function with local properties and RBF kernel function with global properties. For feature extraction, global point density and component structure are fused with local features of pseudo 2D elastic mesh and local point density. Experiment results show the good robustness and high recognition accuracy of the proposed method.