Multi Pattern Search Algorithm (MPSA) for Motion Estimation

Abstract

Estimation of moving objects in image sequence is a central matter in video code, as well as for a variety of tasks in image analysis, motion estimation is exploit the sequential correlation existing in video sequences. Block matching algorithm (BMA) is a technique used for estimating motion vectors in image sequence. It is manipulation divide each frame into block (rectangular or square) then according to spatial search strategy compare between block in reference farm and current frame. Proposed a new motion estimation algorithm multi-pattern search algorithm (MPSA) by considering companied matching criteria and exploit the relationship between the motion models and frames difference of each block, according to motion model choosing suitable pattern, search pattern changing the search area shape to improve motion estimation, while reducing computational complexity. The proposed algorithm is compared with common, simple implemented one which is a full search (FS) algorithm it requires very heavy computational complexity, three step algorithm (TSS), TSS traps into local minimum also these method sacrifice accurate motion estimation to reduce computation. The results show that the proposed algorithm is better according to the computational complexity and accuracy. MPSA check less number of pixel less than FS in (70%) and (30.18%) than TSS and less run time than FS in (65.62%) and (50%) than TSS.