Face Image Recognition Combining Holistic and Local Features

This paper introduces a method using the holistic and the local features for face image recognition. The holistic feature is extracted from spatial domain by 2DPCA and the local feature is taken from 2D-DCT-frequency domain by 2DNMF, respectively. 2D-DCT coefficients form the different frequency components and and get energy concentrate at the same time, which may be suitable to preserve some useful puny features often ignored in global method. And it may avoid the correlation between global and local features and offer complementary frequency information to spatial one. Finally, LSSVM regression is used to weight the mixed feature vectors and classify images. Experimental results have demonstrated the validity of the new method, Which outperforms the conventional 2D-based PCA and NMF methods oil ORL and JAFFE face databases.