Hybrid parallel motion estimation architecture based on fast Pel-subsampling algorithm

In this paper, an algorithm and architecture co-design for parallel motion estimator based on fast pel-subsampling algorithm is proposed. Firstly, we proposed a hard-wareoriented pel-subsampling algorithm to reduce the computational amount of the sum absolute difference (SAD). Moreover, the four-parallel spiral scanning (4PSP) is also utilized to easily perform four candidate blocks. Therefore, the proposed fast algorithm can avoid trapping into the local minimum and save the computational power with a little performance degradation. In the light of our algorithm, the 4times4 processing element (PE) array and the dual mode SAD tree are proposed to perform SAD and SAD with the pel-subsampling (Sub-SAD) efficiently. For the sake of reducing the system memory bandwidth, an efficient memory configuration is also proposed to arrange the current data and reference pixels, and it can achieve the Level C data-reuse scheme.