The optimal transform of motion-compensated frame difference images in a hybrid coder

Using the statistical model proposed by the authors (1992), various statistical characteristics of motion-compensated frame difference (MCFD) images are derived. It is shown that the optimum Karhunen-Loeve (KLT) or the MCFD images is in fact identical to that of the original image sequence in intraframe mode, and that the discrete cosine transform (DCT) remains near optimal. The study this provides insights into the working of the MPEG and H.261 coders, including the justification for using the same DCT for both intra and interframe modes. Experiments on standard image sequences confirm the accuracy of the statistical model.< >