Research on the selection of kernel function in SVM based facial expression recognition

Support vector machine(SVM) means that structural risk minimization principle is used to substitute Empirical risk minimization principle. SVM has shown the excellent performance in pattern recognition. The kernel function is the core of SVM, with which SVM can help to resolve many kinds of non-linear classification problems. Different kernel models and parameters have different result in the performance of the facial expression recognition system. The authors analyze the capability of polynomial kernel function and RBF kernel function in the facial expression recognition using the JAFFE expressions library. The work is valuable in the choise of kernel and its parameters in practice.