Block motion estimation based on spatio-temporal correlation

In this paper, we present a new motion estimation algorithm which determines an initial motion vector and defines a search window adaptively using the motion information of spatio-temporal neighborhood blocks. First, an initial motion vector based on spatio-temporal correlation of motion is predicted. Next, we define a search window adaptively using the Gibbs/Markov random field model. The proposed algorithm particularly deals with the fast moving objects with only a small amount of computation in motion estimation, and is applied to constant and adaptive search windows respectively. The experimental results show the effectiveness of the proposed method