Fast Algorithm for Sub-Pixel Motion Estimation Using Parabola Model for H.264/MPEG4 AVC

In this paper, a fast sub-pixel motion estimation algorithm is presented for H.264/MPEG4 AVC video coding. The proposed methods are based on mathematical models of the motion compensated error distribution function to predict the possible searching direction. Here we do not select the parabola in the horizontal direction but the paraboloid in two dimension by five integer-pixels, the parabola can more exactly describe the moving and texture characteristic in the compressed pictures without strict limitation on the very low bit rate coding. In order to decide the coefficients of the paraboloid model which is suitable for most of sequences, the minimum mean square error (MMSE) from the best integer-pixel motion vector and its horizontally adjacent two half-pixel motion vectors are computed, then the MMSE correlation among three elements are compared and analyzed respectively. By an accurate modeling, we can effectively predict the further searching mode instead of full search. For the quarter-pixel accuracy motion vector, an approximation method is used to get the results from the half-pixel middle results directly. Experimental results show that the proposed method reduces the computational complexity down to 54% averagely, similar performance compared with other fast algorithms in speed, and degradation in the reconstructed video quality is negligible. At the same time, our proposed method does not depend on the integer-pixel searching mode