Hybrid search algorithm for block motion estimation

A hybrid block matching algorithm for motion estimation is proposed in this paper. In this algorithm, a motion block is classified as either large or small motion by making use of the correlation between the block distortion measure of the origin checking point and the magnitude of the motion vector. According to the classified motion type, one of the two fast block matching algorithms, the four-step search and block-based gradient descent search, is employed for the motion estimation. An adaptively updated threshold is used for the motion type classification. Experimental results show that the proposed algorithm can achieve better MSE performance than the two algorithms due to the increased chance for the individual algorithm to search its favorite motion type