Rate-distortion Optimization Based Rate Control

Video compression has made great progress in recent years. With more and more applications, video coding moves beyond mere compression. A joint source-networking coding will provide the best solutions for these applications. One of the key issues for the joint source-networking coding is the bit rate adaptation of the coding system. Rate control is thus a necessary part of an encoder. An encoder employs rate control as a way to regulate varying bit rate characteristics of the coded bit stream to produce high quality decoded pictures at bit rates that are provided by the network. To meet these requirements, we have a deep study on the rate control algorithms for H.264/AVC.Most of recent rate control algorithms, including the algorithms recommended by the international video coding standards, are based on the rate-distortion models. MPEG-2 Test Model Version 5 (TM 5) rate control algorithm is based on the first-order R-D model, while H.263 TMN8, MPEG-4 VM8 and H.264/AVC rate control algorithms are based on the quadratic R-D models. H.264/AVC rate control algorithm doesn't make the same success as MPEG-4 VM8. In this paper, we first take a closer look at these algorithms and illustrate the rationale of the Rate-Distortion models, which are valuable for the design of new rate control algorithms in this thesis.To overcome the limitation of the existing rate control schemes in case of scene changes, we introduce an enhanced linear model for predicting MAD. We proposed a new method to describe the speed of picture complexity changes, by utilizing some knowledge of current frame complexity. Regulated by the new method, predicting MAD turns to be more accurate. Extensive simulation results show that our method improves the video subjective quality significantly, especially when scene changes and high motions occur, and reduces video quality variations considerably in comparison with H.264/AVC rate control algorithms.For fixed-size storage digital video applications, we proposed a well-designed VBR rate control algorithms with almost constant visual quality. With considerations of high visual quality but little real time limitation, a multi-pass rate control algorithm is proposed. In a first pass, the video sequence is encoded with the improved CBR rate control algorithm proposed in the paper, while statistics concerning coding complexity are gathered. In a second pass, the gathered statistics is processed to re-compute the quantization parameters for each frame. The quantization parameters for macroblocks are readjusted according to HVS and coding complexity. The simulation results demonstrate that our method improves the video subjective quality significantly, especially when scene changes occur, and reduces video quality variations considerably.