Fast Sub-pixel Prediction Based on Error Surface Fitting for HEVC

Abstract

The sub-pixel motion estimation is a very important module for improving the motion estimation accuracy in the HEVC video codec standard, which is highly computationally intensive and very time consuming. With the maturity of fast integer-pixel motion estimation algorithm, the time cost of sub-pixel motion estimation becomes more and more significant, so it is very important to develop fast sub-pixel motion estimation algorithm. In this paper, an innovative fast algorithm is proposed for the sub-pixel motion estimation. The algorithm uses the 6-parameter quadratic function to fit the residual surface function, and the matching residuals of the optimal integer pixel prediction point obtained by the whole pixel motion estimation and its adjacent 8 integer pixel points to determine the coefficient of the function, and then the extreme point of the function is obtained, the types of extremum points are discussed, and finally different 1/4 pixel search strategies are applied to the extreme points of different classes. The experimental results show that the algorithm reduces the average number of sub-pixel points to 5 points, greatly reduces the sub-pixel prediction time and computational complexity, while the image coding quality is basically unchanged.