Fast binary motion estimation for screen content video coding

Ting Sun, Pengfei Wan, Oscar C. Au, Wei Dai, Luheng Jia, Yuan Yuan, Amin Zheng, Rui Ma

Abstract: One-bit transform (1BT), followed by binary motion estimation, is an effective alternative for accelerating traditional 8-bit motion estimation (ME) in video coding. The underlining assumption in the design of 1BT methods is that natural videos contain noise. For screen content videos, however, the special characteristics (e.g. screen content is typically noise-free) can be exploited to further improve the motion estimation accuracy. In this paper we propose a binary ME method which is specifically designed for screen content videos. In particular, binary ME is performed on a selected bit-plane. Our contribution is twofold: 1) our bit-plane selection is hardware-friendly and content-adaptive to image blocks; 2) a zero-bias early termination scheme is proposed to accelerates the binary ME procedure. Experimental results demonstrate that our proposed binary ME method improves the accuracy of obtained motion vectors for screen content video sequences.