Two Fast Motion Estimation Algorithms by Using Spatial & Temporal Correlation

Abstract

A good motion estimation algorithm shall use the historic information to provide hints on where to begin the search process. To obtain high performance the properties of motion vectors are studied. From these studies of motion vectors, it is found that the motion vector lengths are short for most cases. Based on the studies, two algorithms for fast motion estimation are proposed. Experimental results show that they achieve significant saving on computation time. To particular, are as fast as the 3-steps search method the reconstructed results are nearly the same as that by the full search method.