Research Of Face Recognition Algorithm Based On Feature Representation

With the development of the related theory of pattern recognition,products with face recognition have been gradually come into the market.But there are all kinds of uncontrollable conditions such as illumination,expression,postureand so on in the application,which would degrade the recognition system’s better performance under controllable conditions sharply.However,in the whole face recognition systems,facial feature representation is a key link in the process of the whole recognition system performance.In the paper,starting from the classification algorith of facial feature representation,one is based on local feature representation,the other is based on global feature representation,principle and steps of several classical algorithms belonging to the two kinds of algorithm are introduced,such as PCA,LDA,LBP and Gabor.The reasearch of facial feature representation is on using preprocessing and fusion algorithm.The main research results are obtained:(1)Because the illumination preprocessing algorithm combined with the facial feature representation can bettersolve the problem of recognition under illumination,an improved retina preprocessing algorithm is proposed in the paper to improve the LBP feature by comparing several facial preprocessing algorithms,in order to improve the LBP feature’s recognition rate under illumination and robustness.The experimental results show the proposed algorithm can improve the performance of facial LBP feature representation.(2)For the single LPQ only using the amplitude information in the frequency domain to represent facial feature,which is relatively weak.A facial feature representation method of local phase quantification based on Monogenic information is proposed by study on the principles of the Monogenic filtering and LPQ operator.The experimental results show the proposed algorithm is with better performance and has the robustness for accessory,expression,age and so on.(3)For the LBP that the pattern of oriented edge magnitudes algorithm used only considering the neighborhood pixels, a new facial feature representation method based on scaled-block LBP of oriented edge magnitude is proposed by using the scaled-block LBP to extract texture feature of gradient accumulative magnitude image.The experimental results show that the improved method improved the performance compared withPOEM algorithm.