Partial face recognition based on sparse non-local regularization weighting

Although the unobstructed face recognition technology has gradually become mature, some face recognition technologies are still in the initial stage, and there are still many difficulties. In the current epidemic situation, people wear masks when they travel. Therefore, the acquired face image is occluded. Because of this, the acquired face image is incomplete, which will lead to insufficient prior information and greatly reduce the face recognition rate. To this end, a partial face recognition method based on sparse non-local regularization weighted coding is proposed. The algorithm uses a full convolutional neural network to obtain facial features for the variability of partial face images. The image of any size is trained and learned, and the face alignment operation of the original image is not needed, and the face information is better preserved. According to the characteristics of local sparse and non-local self-similarity, non-local regularization weighted coding is introduced in the sparse representation classifier. The experimental results show that the proposed method not only solves the arbitrary block of face image recognition, but also has higher recognition and better robustness than the existing algorithms of dynamic feature matching. (Abstract)