| Video coding has been widely investigated for decades due to its importance in many practical applications such as the transmission of high definition TV,internet video,the storage of video signals onto Blu-ray discs,video conferencing and video surveillance.Compared with the last video coding standard H.264/AVC,the newest High Efficiency Video Coding(HEVC)standard achieves about 2x improvement in compression efficiency.However,various video applications have generated huge amounts of video data.Especially in recent years,people have been pursuing more clear,smooth,realistic,vivid and immersive visual experience,leading to explosive growth of video data.Under such circumstances,video coding techniques of higher coding performance is highly desired.Rate-distortion optimization(RDO),one of the most important components in HEVC,plays a crucial role in maximizing coding efficiency.In the current block-based hybrid video coding such as H.264/AVC and HEVC,the RDO is typically performed on the block level individually and independently,which is far from being globally optimal as it ignores the spatial/temporal/inter-view dependency among the block/frame/view coding optimizations and the mutual effect between the components in the hybrid video coding.However,due to the complicated relationships among blocks,a globally optimal solution is extremely hard to be obtained in an operational way and for decades,the global RDO has been open with relatively slow progress.Some typical methods are of high complexity or without comprehensive analysis,making them not efficient for real-time video coding schemes.In this dissertation,considering the temporal dependency in HEVC,a global RDO scheme is proposed which can be implemented in one pass coding.Furthermore,the proposed method is extended for light field integral image(true 3D source)coding.The main contributions are summarized as follows.1.The temporal relationship is examined in the hierarchical coding structure(HCS)in HEVC.To fulfill diversified requirements,HEVC supports several types of HCS configurations such as low delay(LD)and random access(RA).The temporal relationship under these two HCSs are investigated and described with the reference utilization rates.With this temporal relationship,the temporal related units in these two HCSs are identified and the temporal propagation chains are constructed under uni-and bi-prediction with multiple references.2.A temporally dependent RDO(TD-RDO)method is proposed for LD-HCS.Based on the temporal relationship under LD-HCS,the TD-RDO problem is formulated as a distortion propagation problem.Accordingly,a source distortion temporal propogation(SDTP)model is developed for uni-prediction with multiple references,which is used to estimate the degree of distortion propagation and obtain the aggregated propagation.The proposed method is implemented as an adaptation of the global Lagrange multiplier,which is in the form of the independent RDO and thus easy to be implemented.Experimental results show that our proposed method can achieve 2.5% and 2.3% BD-rate gain in average against HEVC under LD-HCS of P and B frames,respectively,with a negligible increase in encoding time.Furthermore,with QP adaptation,higher coding performance can be obtained.3.A RA-based TD-RDO method is proposed.Based on the temporal relationship under RA-HCS,the RA-based TD-RDO problem is decomposed into a layer-wise form,where the temporally related units are then decoupled under bi-prediction.The SDTP model is further developed considering different prediction types including the uni-and bi-prediction with multiple references.The representation of the global Lagrange multiplier for RA-based TD-RDO is obtained based on the formulation of the highest temporal layer under RA-HCS.The effect of introducing random access point picture is thoroughly investigated in the RA-based TD-RDO scheme,including both the distortion propagation analysis and its effect on the distortion propagation of B frames.The proposed method can achieve 2.2% BD-rate saving in average over HEVC under RA-HCS.Our proposed method is also extended for different GOP sizes.4.An inter-view dependent RDO method under 2D-HCS is proposed for light field integral image(extracted as 2D array of sub views)coding.For the sub-view sequence,the inter-view dependency in one GOP and between GOPs is examined to obtain the influence propagation relationship under 2D-HCS.Then based on this relationship,the inter-view rate dependency analysis for bi-prediction is developed under high-rate assumption,which circumvents the issue of distortion propagation,thus greatly simplifying the dependent RDO for light field integral image coding. |
PDF Full Text Request