Face recognition based on Haar LBP histogram

In this paper, a novel face representation approach, Haar Local Binary Pattern histogram (HLBPH), is proposed to represent the face images. First, the face image is decomposed into four-channel subimages in frequency domain by Haar wavelet transform, and then the LBP operator is applied on each subimage to extract the face features. After that, a novel Haar LBP representation method (HLBPH) based on two-layer weighted fusion scheme is presented to balance the face block regions for LBP, to fuse the multi-channel face features. In recognition stage, the Chi square statistic(_χ2) is employed as the dissimilarity measure for face histograms represented by HLBPH. Finally, the proposed algorithm is tested on ORL and Yale face database, and achieve satisfying results to pose, expression and illumination variations.