Face Recognition Based on Wavelet Transform and LBP in Logarithm

In order to improve the performance of face system under complex illumination conditions, we presents a algorithm that combining wavelet transform and LBP (Local Binary Pattern) descriptor in the logarithmic domain to identify face images under complex lighting. First of all, the face images were transformed from spatial domain to frequency domain, and then two level two-dimension discrete wavelet decomposition was calculated on the transformed face images. Then six high frequency coefficients of wavelet transform were used to reconstruct images for face illumination compensation, that is making high-pass filter on face images, elminating the low-frequency lighting componets.Then we partitioned the lighting compensated face images into a grid of fixed size cells, and LBP descriptor extracted the local texture features of each subimages.The features of these subimages were concentration into one feature.At last, we used these features to face recognition. Experimental results based on Yale-B and CMU-PIE show that the proposed algorithm is rubost to complex illumination, and it can express useful features of face images under complex lighting correctly.Experimental results based on Yale-B shows that the proposed algorithm is rubost to complex illumination,and it can express the features of face images under complex lighting correctly.