Novel block matching algorithm for motion vector estimation

Abstract

Motion vector estimation IS the most computationally expensive activity In video compression. Researchers have been trying to develop efficient searching algorithms for long time. In this thesis, we have developed a novel searching algorithm which performs better than the existing popular algorithms in terms of both computational complexity and video quality. To find out the motion vector of a block in the current frame, we utilize the motion vector of the corresponding block at the previous frame, which potentially helps us to jump to an area where the probability of finding a close match is high. After this step, we apply some searching patterns that are capable of moving in all the directions in the searching area using less number of block comparisons. The perfonnance of the proposed algorithm for standard video sequences has been studied and compared with several other popular algorithms. The experimental results show that the number of block comparisons is significantly reduced in our algorithm while the quality of video sequences remains as good as that of the others.