Salih Burak Gokturk Applying 3d Methods to Video for Compression Ee 392j Final Report

Video compression has been an active research area in the last decade. Amongst many proposed methods, motion compensated coding has taken the most attention and taken its place in many standards (Mpeg, H.26L, etc...) The main difference between motion compensated coding and image based coding schemes such as DCT is that motion compensated coding exploits temporal redundancies between consecutive frames, i.e., between two frames. In this paper, we investigate 3D methods that exploit the temporal redundancies on more than two frames. We first describe and evaluate transform based methods such as 3D DCT, and PCA. Next, we propose a new method called volume based motion compensated coding. Volume based motion compensated coding is an extension of regular motion compensated coding and works on stacks of images as opposed to images. Our results indicate that, 3D methods can provide better results than motion compensated coding in many of the video sequences.