Face Recognition Method Via Locally Synthesis Front Face And Sparse Representation

Automatic face recognition has a wide application prospect and has been a hot topic among scholars around the world. Although face recognition has made a great progress, it is still faced with a lot of challenges such as occlusion, expression and illumination. Especially when the face pose varies dramatically, recognition rate will decline. To address this problem, in this paper, we focus on multi-pose face recognition. The main contributions are summarized as follows:A local frontal face synthesizing algorithm is proposed to solve the problem of pose variation in face recognition. In practical face recognition system, the difference between the pose of face for test sample and that of training sample is always huge. In order to reduce errors due to pose variation, we propose frontal face synthesizing algorithm, which can be used to get the virtual frontal face of the profile one. In addition, this algorithm is also dealt with occlusion.We propose a three-phase test sample representation algorithm, which is robust to occlusion, noise and pose. The first phase, frontal face synthesizing phase, synthesizes a frontal face with small horizontal deflection angle using the proposed frontal face synthesizing algorithm. Thus, a frontal face is synthesized as a new test sample. The second phase, training sample selecting phase, selects M training samples that make the most contribution to the representation of the new test sample from all the traning samples. The third phase, decision and recognition phase, recognizes face using the selected M training samples.Experiments on some publicly available face recognition benchmarks demonstrate that the proposed algorithms outperform the state-of-art methods.