A new fast motion estimation algorithm based on search window sub-sampling and object boundary pixel block matching

Abstract

Block motion estimation using full search (FS) is computationally intensive. As a result, fast motion estimation algorithms, such as three-step-search (TSS) and 2-D log search, have been developed. These fast algorithms try to reduce the computational complexity by limiting the number of locations to be searched, but result in higher mean square error (MSE). In this paper, another fast algorithm is introduced, which shows a better MSE performance and lower computational complexity than TSS. The proposed algorithm employs both search window sub-sampling and object boundary pixel decimation techniques to effectively reduce the computational complexity, while keeping even lower MSE at the same time. There are two versions of the proposed algorithm, providing different computational complexity. In particular, the proposed algorithm requires only 32% to 65% of TSS's computation, or in other words, 3.6% to 7.2% of FS's computation, and the performance is in the mid-way between TSS and FS.