Abstract: The whole face matching method includes three steps. First, a face is divided into several regions by using face structure-based region partitioning method; then the face matching process is performed in these regions by using elastic graph with different density and deforming scale, and the optimal matching is found by deterministic annealing algorithm in every region; moreover, the execution of recognition tasks are distributed in grid nodes by using time-constrained minimal cost scheduling algorithm. The experimental results show that our method has higher recognition accuracy and less time than that of other face recognition systems under the large face database and unrestricted conditions.