Face Recognition by Combining Complementary Matchings of Single Image and Sequential Images

This paper describes a face recognition method based on combining two complementary matching algorithms, one is a single-image matching algorithm and the other is a sequential-image matching algorithm. The sequential-image matching algorithm distinguishes each person by using the derived features from a set of sequential images of the same person, and the derived features meaningfully corresponds to the unique face variation pattern of this person. For computing the matching scores, the sequential-image matching algorithm uses a Constrained Mutual Subspace Method (CMSM), and the single-image matching algorithm uses an Euclidean distance. The two matching scores are further combined together so that highly accurate recognition can be achieved. To obtain more robust features under various illuminations, an Anisotropic Smoothing Transform (AST) is also proposed in this paper which can effectively compensates a non-uniform face image into a uniform face image. Experimental results show that the proposed method can achieve outstanding performance and is considerably robust to lighting variations.