Face Recognition Based on Error Detection under Partial Occlusion

Face recognition presents the difficulty of occlusion. The occlusion is generally endowed a little weight to weaken its influence on recognition performance. On the basis of this idea, many existing algorithms used the reconstruction error or projection error as the probability estimation for occlusion image. These methods require iterative computation, which may lead to the difficulty of threshold selection and high time complexity. To solve these problems, this paper proposed a novel method for occlusion face recognition by using an error detection method. First, a face image is divided into four regions and we extract feature and detect error for each region. Second, we use the logarithmic transform error operator to calculate the weight value of each region. The experiments based on the AR database demonstrate that the proposed algorithm for occlusion face recognition achieves high efficiency and good robustness and outperforms the existing methods for certain occlusion recognition.