Research On Some Fundamental Algorithms In Scalable Video Coding

With the rapid development of Internet and multimedia technologies, video streaming over Internet has drawn tremendous attention from researchers of digital video communication. Scalable video coding produces bitstreams decodable at different qualities, frame rates, resolutions and even complexities, and hence is regarded as a promising video coding scheme in the Internet scenario. However, state-of-the-art scalable video coding suffers great performance loss compared with non-scalable video coding, and can only be applied on rectangular video format, which restricts wide application of scalable video coding. This thesis is motivated by these two problems. Based on the current advanced Fine Granularity Scalable (FGS) video coding, the thesis proposed a range-based rate-distortion optimized coder control algorithm together with the generalized trigonometric transform algorithm in the triangular domain. The former can significantly improve the performance of scalable video coding; as a fundamental algorithm in digital signal processing, the latter can provide an efficient candidate transform algorithm not only for irregular domain scalable video coding, but also for other applications in irregular domains, such as pattern recognition, image and graphics processing, geometrical modeling, and so on. The major contributions of this thesis are detailed as follows:Firstly, a two-point rate-distortion optimized coder control algorithm is proposed for advanced FGS video coding. The problem of range-based rate-distortion optimiza-tion(RDO) in scalable video coding is modelled and simplified into a two-point RDO problem for advanced FGS. By investigating the frame dependency in the advanced FGS, this thesis novelly proposes the EOD function to approximate the influence of the parameter selection in the current frame on the following frames. As an example, the EOD model for PFGS(Progressive FGS) is derived, and the two-point rate-distortion optimized (TP-RDO) coder control algorithm is constructed based on this model.Secondly, a two-point rate-distortion optimized joint base layer and enhancement layer mode decision algorithm is proposed for PFGS. The TP-RDO coder control al-gorithm is applied in mode decision of PFGS and three typical weighting, strategies are systematically discussed. Experimental results show that the proposed algorithm can significantly improve the coding efficiency over almost all the rate range. By selecting different weighting strategies, algorithm can flexibly bias to the performance in the low rate range or high rate range.Thirdly, a set of generalized trigonometric functions in the triangular domain are constructed. DCT is the core algorithm in many video coding schemes, while in triangular domain there is no corresponding transform algorithm. By solving the Sturm-Liouville eigen-equation in the barycentric coordinate, the generalized sine and cosine functions in the triangular domain are constructed. Properties of these functions are systematically investigated by visualization and theoretical derivation.Fourthly, discrete generalized trigonometric transform in the triangular domain and related fast algorithms are proposed. Based on the generalized sine and cosine functions in the triangular domain, the discrete generalized sine and cosine transform are defined. Further more, related fast algorithms are constructed by auxiliary functions and transforms. Based on the proposed algorithms, a Matlab subroutine library is implemented. Experimental results show that the discrete generalized cosine transform is helpful for transform coding in triangular domain.