Adaptive motion vector resolution based on rate-distortion cost and coding unit depth

Abstract

In current HEVC, motion vector resolution is always fixed to 1/4 pixel for the whole video sequence. Inter coding with fixed motion vector resolution can decrease coding efficiency because statistical property of the local image is not considered. The prediction errors and the number of texture bits decrease by the motion compensation, but the number of coding bits of the motion vector increases for the increased motion vector resolution. In contrast, when motion vector resolution decrease, the number of coding bits of the motion vector decrease, but the prediction errors and the number of texture bits increase. In this paper, we proposed adaptive decision scheme for motion vector resolution to improve the coding efficiency. This contribution proposes an adaptive mechanism for threshold selection at a PU level in the Progressive Motion Vector Resolution(PMVR) method. The proposed scheme capitalizes on the tendency that higher pel-precision is more beneficial in terms of coding efficiency as the CU depth decreases. Also, strength determined based on the RD cost and select predefined threshold set per each slice in order to consider the local characteristics of the image for improving coding gain by avoiding unnecessary motion vector search and minimizing the additional transmission bits. Simulation results with respect to HM7.0 show that the proposed scheme provides coding gain of 3.4% for Low delay P structure, 1.4% for Low delay B structure. Moreover, it is reported that average encoding time is saved by 5%. Also, proposed scheme can improve coding efficiency at slightly increased encoding time compared to PMVR method.