Face recognition based on weighted multi-resolution kernel entropy component analysis
Xiaoli Ruan, Shunfang Wang · 2017
In order to overcome the instability of non-linear high dimensional facial data caused by illumination and expression, and obtain rich, effective and complementary face features, we put forward a method for face recognition by improving kernel entropy component analysis (KECA) with weighted multi-resolution. First, we use traversal algorithm to search an optimal weight for the multi-resolution face feature, and extract the features of each face feature set via Gabor wavelets. Second, we utilize the nonlinear dimension reduction algorithms of KECA and KPCA (kernel principal component analysis) to process the data, respectively, and compare the results of these two. Third, K-nearest neighbor is used for final classification on the fusion of different weighted multi-resolution human face images. The experimental results of the ORL and YELA face database showed that our proposed method has a high recognition ability and stability compared with the common feature reduction algorithms.