Abstract

The focus of this thesis is the design of rate-control (RC) algorithms for constant quality (CQ) video encoding and transcoding, where CQ is measured by the variance of quality in PSNR (peak signal-to-noise ratio). By modeling DCT coefficients as having Laplacian distributions, Laplacian rate/models are developed for MPEG-4 encoding and transcoding. These models accurately estimate the rate and distortion (in PSNR) of MPEG-4 compressed bitstreams. The rate model is applied to a CBR (constant bit rate) encoding algorithm. This algorithm offers a better or similar PSNR as compared to the Q2 [7] algorithm with a lower variation in bitrate. Thus, it outperforms Q2. These models are then applied to CQ video coding and transcoding. Most CBR control algorithms aim to produce a bitstream that meets a certain bitrate with the highest quality. Due to the non-stationary nature of video sequences, the quality of the compressed sequence changes over time, which is not desirable to end-users. To provide a solution to this problem, six CQ encoding algorithms are proposed: the first two are VBR (variable bit rate) algorithms with a fixed target quality (FTQ), the next two are CBR algorithms with FTQ, and the last two are CBR algorithms with a dynamic target quality (DTQ). Within each group of two, the quality is controlled either at the frame level (using the Laplacian rate/distortion model) or at the macroblock level (using the actual distortions). With the success of these algorithms, the CQ DTQ encoding algorithms are extended to MPEG-4 video transcoding (bitrate reduction with requantization). These CQ transcoding algorithms can handle the problems that are uniquely present in transcoders, such as the lack of the original sequence and requantization. Similar to their encoding counterparts, these CQ transcoding algorithms have an extra degree of freedom to balance the quality variation with the accuracy to the target bitrate and the average quality. Simulation results indicate that these algorithms offer lower PSNR variance while having similar/lower average PSNR and bitrate when compared with Q2T and TM5T (transcoding version of Q2 and TM5). Besides proposing MPEG-4 CQ RC algorithms, an MPEG-2 rate-control algorithm is also developed based on TM5. It aims at improving the subjective quality measured by using Watson's DVQ (digital video quality) metric. When compared with TM5, it provides a better DVQ. However, since Watson's DVQ metric is not a standard way to estimate the subjective quality, PSNR is still used in the rest of the research