Abstract: In view of the situation of low recognition rate and bad robustness of the face recognition based on the single modal features and smart environment, a new multi-feature fusion face recognition method was proposed based on global and local feature fusion and kernel relevance weighted discriminant analysis(KRWDA) algorithm. In order to solve the problem caused by small sample size, a new algorithm called KRWDA was proposed by combination of the relevance weighted discriminant analysis and the kernel trick. The global features and local features were fused in feature layer using the complementarity of the contents and functions described at recognition, and the global features and the local features were extracted by discrete cosine transform and Gabor wavelet transform respectively. The simulation experiments on database AMIES2016 indicated that the proposed method can enhance the accuracy of identity recognition system effectively.

Key words: computer application, feature fusion, global feature, local feature, kernel relevance weighted discriminant analysis(KRWDA) algorithm