| Summary: | Video compression has numerous applications in modern multimedia communications and storage such as digital television (DTV), Internet video streaming and digital versatile disk (DVD) video storage. The strong correlations between successive video frames and within individual frames make it possible to achieve high compression by reducing the temporal and spatial redundancies. Without video compression, all the digital video transmission and storage would not be practical. Various video compression standards (MPEG-1/2/4, H.261/263 and H.264/AVC etc.) have been developed since early 90’s and they have been widely used throughout the years. Although these standards have specified the decompression procedure, many issues regarding the implementations of compression algorithms are open for optimization. In this thesis, we study various fundamental techniques in motion compensated video compression including motion estimation, mode decision and encoder rate control. The goal of our research is to achieve better tradeoff between encoding complexity and quality as well as better utilization of the available channel bandwidth.
We first investigate block matching based motion estimation techniques. In video coding, motion estimation is the most computation intensive task and therefore to achieve significant complexity reduction in motion estimation without reducing too much the estimation quality is highly desirable. We propose two fast algorithms based on the adaptive search pattern and partial distortion metric respectively. Both algorithms explore the statistical characteristics of motion vector distribution and can adaptively select suitable search points or block matching criteria for different areas within the search window. Experimental results show that, by using the proposed fast approaches, the motion estimation complexity is greatly reduced with only slight degradation in estimation accuracy. Moreover, we propose a new measure of similarity between blocks in motion estimation called the smooth constrained mean absolute difference (SCMAD). Compared with the traditional MAD criterion, we achieve reduced bit rate for encoding the residue block without any degradation of the reconstructed image quality.
Variable size block motion estimation is a very important technique for video coding. The latest H.264/AVC standard employs seven different size block types in motion estimation which can significantly improve the coding performance compared with the previous video coding standards. In order to substantially reduce the computational complexity of the motion estimation and mode decision process in H.264/AVC, we introduce an efficient inter mode decision algorithm based on the macroblock (MB) classification. The proposed technique can save the computation cost significantly while maintaining the same rate-distortion (R-D) performance.
The main objective of rate control is to regulate the output bit rates of video encoders to achieve the best utilization of the available channel bandwidth. We propose two frame complexity based schemes for H.264/AVC video rate control. The first scheme is an adaptive intra-frame rate-quantization (R-Q) model using gradient-based frame complexity measure. This model aims at selecting accurate quantization parameters (QP) for intra-coded frames. In addition, we develop a frame level target bits allocation scheme which takes into consideration both the buffer status and the predicted frame MAD value. We show that the overall performance of the rate control, in terms of target bits mismatch and the visual quality of reconstructed video, can be improved.
|
|---|