Adaptive rate control for advanced video coding

Rate control serves as an important technique to moderate the bit rate of video transmission over a channel of limited bandwidth and to control the bit allocations within a video sequence to maximize its overall visual quality. The output bit rate and video quality of a video encoder depend on several coding parameters such as the quantization parameter and coding mode. In particular, choosing a large quantization parameter reduces the resulting bit rate, while at the same time reducing the visual quality of the encoded video. Since the level of redundancy changes from frame to frame, the output bit rate is inherently variable. The goal of rate control, therefore, is to keep the output bit rate within constrained limits such that the buffer does not overflow or underflow while achieving maximally uniform video quality. This goal can be achieved by optimal bit allocation and accurate quantization parameter determination. To be more specific, the bit allocation scheme is employed to distribute the bit budget among coding units such as frames, or macroblocks in such a way that the overall distortion is minimized. The accurate quantization parameter for the encoder is then determined to meet the bit budget.; In this dissertation, we propose several techniques to improve rate control performance. First, we propose a unified complexity-based (MAD-based) bit allocation method optimized in rate-distortion sense for both the macroblock layer and frame layer.; To compensate for the inadequacy of the MAD-based method when the motion in a frame becomes fast and complex, we introduce PSNR-based method as an enhancement. In order to deal with scene changes in video sequence, we propose a simple but effective scene change detection method.; Since rate control can be achieved by the methods of adjusting the quantization parameter or controlling the video frame rate (frame skipping), a novel quantizer adjustment scheme and an improved frame skipping scheme are proposed.; To reduce the mismatch between the actual bits and target bits, we propose an adaptive Lagrange multipler in rate-distortion optimization (RDO) process.; Based on these techniques, we then present an overall frame layer rate control algorithm for H.264/AVC. When compared to the existing rate control scheme, it achieves significant performance improvement and better buffer regulation due to its accurate quantization parameter and less mismatch between the actual encoding rate and target bit budget. It also handles scene changes better than conventional approaches.; We also propose an optimized macroblock layer rate control for low-delay applications. It is more challenging as it is required to satisfy the low end-to-end delay constraint, and hence a very small buffer size constraint. By taking into consideration the actual encoding results of previously coded macroblocks, we propose techniques in bit allocation and quantization parameter adjustment to prevent the small buffer from overflow or underflow. Simulations show that our rate control algorithm gives better and more stable quality, especially at lower bit rates where the impact of rate control scheme is more significant. (Abstract shortened by UMI.)...