High definition (HD) video contents become popular and displays of higher resolution such as ultra definition are emerging in recent years. The conventional video coding standards offer excellent coding performance at lower bit-rates. However, their coding performance for HD video contents is not as efficient. The objective of this research is to develop a set of efficient coding tools or techniques to offer a better coding gain for HD video. The following three techniques are studied in this work. ❧ First, we present a Joint first-order-residual/second-order residual (FOR/SOR) coding technique. The FOR/SOR algorithm that incorporates a few advanced coding techniques is proposed for HD video coding. For the FOR coder, the block-based prediction is used to exploit both temporal and spatial correlation in an original frame surface for coding efficiency. However, there still exists structural noise in the prediction residuals. We design an efficient SOR coder to encode the residual image. Block-adaptive bit allocation between the FOR and the SOR coders is developed to enhance the coding performance, which corresponds to selecting two different quantization parameters in the FOR and the SOR coders in different spatial regions. It is shown by experimental results that the proposed FOR/SOR coding algorithm outperforms H.264/AVC significantly in HD video coding with an averaged bit rate saving of 15.6%. ❧ Second, we develop two advanced processing techniques, which are referred as to two-layered transform with sparse representation (TTSR) and slant residual shift (SRS), for prediction residuals so as to improve coding efficiency. Prediction residues often show a non-stationary property, and the DCT becomes sub-optimal and yields undesired artifacts. The proposed TTSR algorithm makes use of sparse representation and is targeted toward the state-of-the-art video coding standard, High Efficiency Video Coding (HEVC), in this work. A dictionary is adaptively trained to contain featured patterns of residual signals so that a high portion of the energy in a structured residual can be efficiently coded with sparse coding. Then, the following DCT in cascade is applied to the remaining signal after spare coding. The use of multiple representations is justified with an R-D analysis, and the two transforms successfully complement each other. The SRS technique is to align dominant prediction residuals of inter-predicted frames with the horizontal or the vertical direction via row-wise or column-wise circular shift before the 2-D DCT. To determine the proper shift of pixels, we classify blocks into several types, each of which is assigned an index number. Then, these indices are sent to the decoder as signaling flags, which can be viewed as the mode information of the SRS technique. It is demonstrated by experimental results that the proposed algorithm outperforms the HEVC. ❧ Third, we make a contribution to the HEVC with several efficient coding tools incorporated into the two context adaptive entropy coding, i.e., Context Adaptive Variable Length Coding (CAVLC) and Context Adaptive Binary Arithmetic Coding (CABAC). The proposed tableless VLC coding scheme removes all the tables used for the residual coding, yet yields negligible changes to the coding performance. Statistical property of a symbol is employed to replace the conventional tables with a mathematical model and improve a coding gain with a high order Markov model. On top of that, a context for a significance map coding in a large transform block is newly designed. The proposed context model removes a dependency of neighbor significant coefficients along the same scanning line, and, thus, it enhances a throughput of the CABAC. The proposed algorithm extends to the mode dependent coefficient scanning method for a large transform block. The proposed algorithm has negligible effect on the coding performance while it significantly improves the parallelization.
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.