MPEG-2 Motion Estimation using a PC Cluster

D. Grosu and H. Gâlmeanu

Keywords

MPEG, motion estimation, motion vector, block matching, macroblock, PVM, partial distance

Abstract

Motion estimation is the most computing-intensive component of the MPEG-2 encoding process. It refers to the action of searching for the closest prediction of a certain image block. The more accurate the prediction, the better the achieved compression and quality are. The optimal solution is guaranteed by performing an exhaustive search for all possible predictions for a certain block. The motion estimation component is characterized by a polynomial time, and is responsible for about 90% of the encoding time. In this article the search time is reduced by parallelizing the block matching search algorithm, used in addition to the partial distance method to prevent useless but time-expensive calculations. Partial distance and pixel spiral summation order are used to reduce the time spent on error summation. Experimental results show that significant improvements in performance can be made. The encoding time using a 128 × 128 window is reduced from 14 sec/frame, in the sequential case, to 3.2 sec/frame, about five times faster, using six Pentium 450 MHz in parallel.

Important Links:



Go Back