With the great advancement of multimedia technologies in the last decades, there has been an enormous surge in the number of multimedia involved applications. Meanwhile, the importance of multimedia technologies is constantly growing. Among all the key techniques, sophisticated multimedia compression algorithms that provide superior coding performances are much desired. After the multimedia is compressed, it needs to be transmitted across networks, with applications including video conferencing, telepresence, direct broadcast satellite (DBS) and video-on-demand (VoD) services. However, due to the bandwidth constraint, the total available bit budget is often limited. Therefore, how to efficiently allocate the total bit budget among the coding units is a critical issue. Moreover, the predi...[ Read more ]

With the great advancement of multimedia technologies in the last decades, there has been an enormous surge in the number of multimedia involved applications. Meanwhile, the importance of multimedia technologies is constantly growing. Among all the key techniques, sophisticated multimedia compression algorithms that provide superior coding performances are much desired. After the multimedia is compressed, it needs to be transmitted across networks, with applications including video conferencing, telepresence, direct broadcast satellite (DBS) and video-on-demand (VoD) services. However, due to the bandwidth constraint, the total available bit budget is often limited. Therefore, how to efficiently allocate the total bit budget among the coding units is a critical issue. Moreover, the predictive coding nature in hybrid video coding makes the problem much more challenging.

In this thesis, we address the problem of dependent bit allocation and rate-distortion modeling in hybrid video coding. First, an analytical framework for frame-level dependent bit allocation in hybrid video coding is proposed. Starting from the nature of the predictive video coding, the dependency of neighboring frames is quantitatively measured with the proposed inter-frame dependency model. The dependent bit allocation problem is revisited. In stead of employing the search-based methods which are very popular in the literature, a model-based bit allocation framework is introduced. The frame-level dependent bit allocation problem is carefully formulated and solved using successive convex approximation techniques. In addition to the coding dependency of inter-frames, the inter-block dependency in intra-frames is discussed. It is well known that for intra-frames, the pixels at the block boundaries usually have larger distortion than the pixels at other positions. This biased distortion distribution not only generates unpleasant blocking artifact, but potentially decreases the intra-prediction performance and eventually diminishes the coding efficiency. Motivated by this observation, a novel distortion redistribution approach is proposed which aims to achieve the best rate-distortion performance. Under the total distortion constraint, the consumed bits are minimized by utilizing the intra-frame coding dependency.

Next, we investigate the dependent joint bit allocation problem in statistical multiplexing systems in which the transmission channel is shared among multiple video programs. Different from the existing method, both the relative frame complexity and the inter-frame dependency are considered in the problem formulation, and the final bit allocation strategy is achieved via convex relaxation techniques.

After the bit allocation, the next key problem is how to determine the coding parameters such that the generated bits would be as close to the allocated bits as possible. To achieve this, accurate rate-quantization and distortion-quantization models need to be determined. Thus, we proposed to use the Laplacian Mixture Model (LMM) to approximate the distribution of the DCT coefficients due to its superior estimation accuracy. It can be seen that the proposed models can handle both standard definition video sequences and high definition video sequences. Based on LMM, the corresponding rate-quantization and distortion-quantization models are developed, and they are used in conventional rate control methods.