High Quality Image and Video Compression for Lossy Packet Networks

摘要

This project addressed compression of images and video using group testing. We implemented a video codec called group testing for video (GTV) that uses the discrete cosine transform (DCT), followed by bit-plane coding of the DCT coefficients. GTV outperforms H.263 at medium-to-high bit rates. We also developed several new rate control algorithms to work with GTV, taking advantage of its ability to assign very precise numbers of bits to each frame to achieve constant, or near-constant, quality per video frame. Constant quality video is important because it means that there are no glitches or objectionable artifacts ill the video. Usually, when a scene change occurs, there is a noticeable drop in quality because more bits should have been allocated to the frames around the scene change, and$or frames could be dropped. We are currently implementing some of the features of the new H.264 video coding standard in GTV. These include multiple-frame motion compensation, variable block-size motion compensation, quarter-pixel motion compensation, and 4X4 integer DCTs. These new features are leading to improved video quality over our initial GTV coder.