Research On Rate Allocation Technology Based On Discrete Wavelet Multi-view Video Coding

With the development of communication and network technology, higher demandis imposed on video services. Due to the better visual experience, a wide variety ofmulti-view video applications are emerging. The current video communication technol-ogy and existing video processing equipments cannot meet the requirement of efcienttransmission. Efcient compression becomes more important in storage and delivery ofmulti-view video content. Rate allocation is one of the most important parts in video cod-ing. Rate allocation aims to achieve a balance between reconstructed video quality andtransmission bandwidth to satisfy special requirements of specific application. Rate allo-cation takes objective video quality, subjective video quality, actual bit rate and specificneeds of application into account. An efective and efcient rate allocation algorithm notonly improves the reconstructed video quality, but also enables more efcient channelsutilization.Wavelet-based video coding is an important video coding method, and it has beenused in multi-view video coding. The encoder allocates the number of bits to each waveletsubbands using the rate-distortion information generated in the encoding process. Therate allocation approach using Lagrange multiplier can be use to minimize the overalldistortion given a fixed rate budget. As the optimization objective is the total distor-tion, this scheme results in quality fluctuations. Video quality fluctuations result in aless aesthetically-pleasing experience for the viewer. Consequently, a rate allocation al-gorithm to control quality fluctuation, or distortion variation, is highly desirable. In thisdissertation, considering the features of rate allocation, three aspects are studied. They arerate-distortion (RD) modeling, rate allocation algorithms and rate allocation adjustmentalgorithms wavelet-based in multi-view video coding (MVC).Firstly, a RD model of wavelet subbands is proposed. Under the assumption that sub-band coefcients obeys multi-dimension Gaussian probability distribution, the RD modelis obtained. Based on the statistical investigation of subband coefcients, a correlationmodel is developed, which is used as a covariance matrix of multi-dimension Gaussianprobability distribution. Covariance matrix diagonalization decorrelates the coefcientand the context. RD models with their respective model coefcients are used to calculatesubband RD and form RD function, which forms the basis of rate allocation. Secondly, the motion-compensated temporal filtering (MCTF) decorrelates the videosequences in the temporal axis and the disparity compensated view filtering (DCVF)decorrelates the video sequences in the view axis. The relation between subband andreconstructed video directly relates to the wavelet transformation aforementioned. Thedistortion of reconstructed video is determined by the distortion of wavelet subbands. Af-ter the motion estimation, the categories of pixels are explored and fractions of categoriesare calculated. Considering the diferent weights of each temporal subband obtained fromnonorthogonal wavelet transform, an error propagation model is built.Finally, reconstructed video quality fluctuation is divided into two types, namely,inter-GOP and intra-GOP, according to reasons cause the fluctuation. For inter-GOP fluc-tuation, a two-pass rate allocation algorithm is proposed, in which the RD model is con-structed to represent the content characteristics of whole GOP in the first-pass procedureand the rate assigned to GOP will be adjusted within the second-pass according to thesemodels, so that video sequences can be coded with a constant quality. In the first-pass, RDmodel is estimated from subband’s RD information and the expected distortion of w-hole GOP under the constraint of the target bit rate. Based on the error propagation model,the aim to control quality fluctuation within GOP is posed as a quadratic programmingproblem and its solution forms the basis for proposed algorithm.