Face recognition using local statistics of gradients and correlations

Most of face recognition methods often use a raw image itself for a feature vector. However, the feature vector directly formed from a raw image is seemed to be susceptible to variation of illumination and facial expression. In this paper, we propose a face recognition method using local statistics of gradients and correlations. BDIP (block difference of inverse probabilities) is chosen as a local statistics of gradients and two types of BVLC (block variation of local correlation coefficients) as local statistics of correlations. When a test image enters the system, it extracts the three types of feature vectors, fuses them, and classifies the image by using whitened PCA process and cosine distance. Experimental results for the three face DBs, Yale, Yale B, and Weizmann, show that the fused features of BDIP and BVLCs are more robust to variation of illumination and facial expression and so the proposed method yields good results.