Motion estimation in 2D-DCT transform domain
Petrescu Catalin-Dumitru, Stefanoiu Dan, Ciprian Lupu · International Conference on Information and Automation · 2008
Motion estimation is one of the most important steps in video compression algorithms. It reduces temporal redundancy present in frame sequences and allows a better compression of video material. Most of the actual video compression algorithms use matching methods which operate on the bitmap form of the frames. Because real motion in video frames is not carried out over integer pixels, sub-pixel motion estimation is performed for improving the quality of the compression. Also, most of the video compression algorithms use discrete cosine transform for coding the residual frame information (in the motion compensation step). For this reason, motion estimation algorithms which operate directly in discrete cosine transform domain are of great interest. This paper presents a method for computing the values of DCT coefficients of a block of pixels positioned on certain coordinates over four adjacent blocks using only the DCT coefficients of these four blocks. Also, an equivalent of the full-search algorithm translated in 2D-DCT domain is presented.