In this paper, we used image sequences to generate a velocity vector field. This field has been used for the face extraction. To evaluate the velocity field, we proposed a modified constraint line clustering algorithm which could generate the field in a very short time. The result from the face segmenation was fitted to an ellipse with second-order spatial moments. The representation of an ellipse enabled us to rotate the face in fast and easy way. We used a template matching scheme to identify each face. In this scheme, we used an intensity, scale invariant algorithm. With this algorithm, we obtained a successful recognition performance. Also, the face recognition algorithm, we proposed, showed to be superiority over all other approaches especially in view of the recognition time.
RV860-PIPE is the pipelined MIMD machine which we developed for computer vision systems. It is based on a ring-topology and consists of four node boards, each with a 64bit RISC, i860 processor. With the proposed algorithm, we implemented the real-time face recognition system, which could indentify a person in two seconds. The reason that we could obtain such a fast recognition speed resulted not only from the processing power of one processor, but also from the effective parallelization.