Predictive motion vector field adaptive search technique (PMVFAST): enhancing block-based motion estimation

TLDR: In this paper, the Predictive Motion Vector Field Adaptive Search Technique (PMVFAST) was proposed to accelerate the full search algorithm by using predictive techniques and early termination criteria, which make use of parameters adapted to local characteristics of a frame.

Abstract: Motion Estimation (ME) is an important part of most video encoding systems, since it could significantly affect the output quality of an encoded sequence. Unfortunately this feature requires a significant part of the encoding time especially when using the straightforward Full Search (FS) algorithm. In this paper a new algorithm is presented named as the Predictive Motion Vector Field Adaptive Search Technique (PMVFAST), which significantly outperforms most if not all other previously proposed algorithms in terms of Speed Up performance. In addition, the output quality of the encoded sequence in terms of PSNR is similar to that of the Full Search algorithm. The proposed algorithm relies mainly upon very robust and reliable predictive techniques and early termination criteria, which make use of parameters adapted to the local characteristics of a frame. Our experiments verify the superiority of the proposed algorithm, not only versus several other well-known fast algorithms, but also in many cases versus even the Full Search algorithm.