Recognizing Partially Occluded Faces from a Single Exemplar Image Per Person

Despite remarkable progress on human face recognition, little attention has been given to robustly recognizing partially occluded faces. In this paper, we propose a new approach to recognize partially occluded faces when only one exemplar image per person is available. In this approach, a face image is represented as an array of Patch PCA (PPCA) extracted from a partitioned face image containing information of local regions instead of holistic information of a face. An adaptive weighting technique is utilized to assign proper weights to PPCA features to adjust the contribution of each local region of a face in terms of the richness of identity information and the likelihood of occlusion in a local region. The encouraging experimental results using AR face database demonstrate that the proposed method provides a new solution to the problem of robustly recognizing partially occluded faces in single model databases.