A Local Difference coding algorithm for face recognition

Stimulated by the popular Local Binary Pattern (LBP) coding, we propose a new operator, i.e. the Local Difference (LD) coding. Different from other local coding approaches, the LD coding extracts direction information of both first-order and second-order difference. We apply the LD coding in feature extraction for face recognition, and obtain a very low dimensional feature in comparison with the classic operator such as the Uniform LBP (U-LBP). Experimental results show that the feature obtained with the LD operator could increase or maintain the recognition rate with much lower number of feature dimension.