Multiple Granularity-Based Feature Combination for Face Recognition

In face recognition, coarsening-granularity image blocks mainly reflect some contour features and global information of human face identity. On the other hand, refining-granularity face parts can carry more local features about identifying information, for example: mouth, eyes and brows, etc. This paper proposes the framework of multi-granularity feature combination. Under the framework, a multi-granularity feature combination algorithm (MGFC) for face recognition is given, which uses the two-scale windows and two-stage feature extraction method. Some experimental results on benchmark face database demonstrate that MGFC algorithm is effective and can obtain a higher recognition rate than single granularity feature extraction algorithm.