Feature selection for pose invariant face recognition

One of the major difficulties in face recognition systems is the in-depth pose variation problem. Most face recognition approaches assume that the pose of the face is known. In this work, we have designed a feature based pose estimation and face recognition system using 2D Gabor wavelets as local feature information. The difference of our system from the existing ones lies in its simplicity and its intelligent sampling of local features. Intelligent feature selection can be carried out by learning a set of parameters where the aim is the optimal performance of the overall system. In this paper we give comparative analysis of the performance of our system with the standard modular eigenfaces approach and show that local feature based approach improved the performance of both pose estimation and face recognition. For efficient coding, we have employed principal component analysis to the outputs of local feature vectors. Intelligent feature selection also reduced the space and time complexity of the system while retaining almost the same estimation and recognition accuracy.