Skin texture features for face recognition

Face recognition has been deployed in a wide range of important applications including surveillance and forensic identification. However, it still seems to be a challenging problem as its performance severely degrades under illumination, pose and expression variations, as well as with occlusions, and aging. In this thesis, we have investigated the use of local facial skin data as a source of biometric information to improve human recognition. Skin texture features have been exploited in three major tasks, which include (i) improving the performance of conventional face recognition systems, (ii) building an adaptive skin-based face recognition system, and (iii) dealing with circumstances when a full view of the face may not be avai'lable. Additionally, a fully automated scheme is presented for localizing eyes and mouth and segmenting four facial regions: forehead, right cheek, left cheek and chin. These four regions are divided into nonoverlapping patches with equal size. A novel skin/non-skin classifier is proposed for detecting patches containing only skin texture and therefore detecting the pure-skin regions. Experiments using the XM2VTS database indicate that the forehead region has the most significant biometric information. The use of forehead texture features improves the rank-l identification of Eigenfaces system from 77.63% to 84.07%. The rank-l identification is equal 93.56% when this region is fused with Kernel Direct Discriminant Analysis algorithm.