Abstract: In order to improve the speed and accuracy of motion estimation in adaptive rood pattern search(ARPS)algorithm， a fast block matching motion estimation algorithm was proposed on the basis of spatial correlation prediction. According to the block matching metric， in the proposed algorithm， the motion vector(MV)of current block was predicted by computing mean MVs of the two blocks with higher similarity in the neighborhood blocks instead of the fixed single block prediction pattern in ARPS. The prediction performance of initial search center location was improved with the proposed algorithm， and the prediction error resulted from motion transformation was reduced. Experimental results demonstrated that the computational complexity could be effectively decreased， and the registration accuracy was increased as well in comparison to APRS and the other standard block matching algorithms.

Key words: motion estimation, video coding, prediction vector, spatial correlation, block matching algorithm