Adaptive Pixel-based Direction Oriented Fast Motion Estimation for Predictive Coding
Tushar Shankar Shinde · 2022
This paper presents a novel Adaptive pixel-based direction-oriented fast motion estimation (APDME) method for lossless video compression. The traditional approach of block-based motion estimation suffers from two major issues: 1) the need to transmit side information such as motion vector overhead, and 2) incorrect motion estimation at object boundaries. The proposed method predicts motion on a pixel-by-pixel basis and addresses both the highlighted issues. In this pursuit, we present three contributions: 1) a novel target window design consisting of a spatially predicted pixel as a part of the search process, 2) initial MV prediction consisting of a global MV candidate, and 3) a novel direction oriented search patterns for faster search convergence. The proposed method achieves better entropy results and a significant reduction in the computations than pixel-based full search for varying motion content test sequences. Experimental results also suggest the proposed approach achieves superior entropy performance for fast and directional motion sequences with a computational edge over other competitive methods.