A fast SURF way for human face recognition with Cell Similarity

Face recognition is a very challenging problem in computer vision. In this paper, Speeded up Robust Features (SURF), a scale and rotation invariant interesting point descriptor, is further explored for face recognition. Specially, a novel technique, Cell Similarity is proposed to make improvement based on SURF in face recognition. In the meantime, different cell division strategies are proposed and evaluated in this paper, which move towards revealing the inner relation and essence in face recognition. We not only obtain good results in ORL dataset and our Lab dataset (aligned face), but also speed up the original version by reducing matching time. Moreover, in order to further deal with rotation situation, another new loopy Cell Similarity method in these two datasets is evaluated, and advantages and disadvantages of different implementations are also discussed.