Robust Face Recognition Based on DCNN and CRC

Collaborative representation based classification (CRC) has gained popularity in recent years, but the conventional features could not effectively handle the variations of pose and occlusion such as sunglass. In this paper, the deep convolution neural network (DCNN) is introduced in the feature extraction to tackle the facial deformation, particularly, such as pose variation. After analysis on the feasibility of the dictionary construction on DCNN based features, a DCNN feature based occlusion dictionary computing algorithm is then presented to tackle the face recognition with occlusion. Experiments on representative face databases with variations of pose and occlusion demonstrated the effectiveness of the proposed algorithm scheme.