Font Size: a A A

Research On Rate Control And Optimization In Video Coding

Posted on:2020-05-30Degree:DoctorType:Dissertation
Country:ChinaCandidate:H W GuoFull Text:PDF
GTID:1368330623958185Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
High efficiency video coding(HEVC)is the latest video coding international standard after H.264/AVC.The standard specifies the decoder and the syntactic structure of the bit stream.But it does not specify the implementation for video coding at the encoder side.Therefore,designers can flexibly develop the coding strategies according to the requirement for the practical application in order to achieve the optimal coding performance under the different target bit rate.Rate-distortion(R-D)optimization is a very important technology in video coding,which is used to select coding parameters and regulate the trade-off between coded distortion and coded bit rate.Rate control is an indispensable module in video coding systems,which makes the encoder output a specific bit rate to meet the requirements of channel bandwidth and storage space.Based on the platform of HEVC encoders,this dissertation studied the both video coding optimization and rate control by analyzing the R-D dependency among coding units.Specific research contents and innovations include the following aspects:(1)In the HEVC encoder,the coding parameters including coding modes,motion vectors,reference frames and quantization transform coefficients,etc.,are independently determined by calculating the minimum R-D cost in a basic coding unit.There are R-D dependencies among coding units due to pixel prediction and motion vector prediction,and so on.Therefore,the coding decision of the current coding unit may affect the maximum R-D performance that the subsequent coding units can achieve.In order to better apply these R-D dependencies to the design of video coding optimization and rate control,this dissertation studied the R-D characteristics and R-D dependencies at different levels of coding units in HEVC.Through theoretical analysis and experimental statistics,it is found that a power function can well represent the operational R-D curves for different levels of coding units.In addition,for the R-D dependencies,the temporal dependency is more significant,while the spatial dependency is not obvious,and the distortion of referenced pixels mainly affect the coded distortion rather than the coded bit rate for the subsequent coding units.(2)By analyzing the R-D dependency between the coding frame and the reference pixels reconstructed,and analyzing the R-D influence intensity of different coding blocks in the current frame to subsequent coding process,a R-D dependency based two-pass encoding optimization method was proposed.In the proposed method,each frame is encoded twice consecutively.First,a frame is encoded by a simplified inter prediction with the original HEVC quantization parameter(QP)setting,for obtaining the influence factors at frame level and block level which are used to measure the R-D intensity.Then,the second encoding is performed after determining the new QPs and Lagrange multipliers according to the influence factors.The experimental results show that the proposed method obtains bit rate savings of 5.1% and 5.3% on average under the configurations of low-delay B and P frames,respectively.In addition,the increased computational cost is only 23% on average.(3)For the problem that traditional rate control methods often reduce the R-D performance of encoders,a frame-level bit allocation and bit rate control based on global optimum criterion was proposed.First,an accurate frame-level R-D model estimation method was proposed by analyzing the low-delay coding structure of HEVC.Second,an improved group of picture(GOP)-level bit allocation method was proposed according to the R-D importance of different GOP in the video sequence.Finally,an optimal frame-level bit allocation equation under the condition of rate control was established according to the global R-D optimization,and the equation was resolved by combining the frame-level R-D model and GOP-level bit budget.The proposed method not only achieves accurate bit rate control,but also significantly improves the R-D performance of encoders.Specifically,compared with the default HEVC encoder without rate control,the proposed method obtains the R-D gains of 3.2% and 3.7% on average under the configuration of low-delay B and P frames,respectively.(4)In order to further improve the accuracy of bit rate control and the R-D performance,an inter-block temporal dependency based coding tree unit(CTU)-level rate control method was proposed.First,the global optimization formula at CTU-level was established based on the R-D optimization theory,which is expressed as a weighted Lagrange multiplier method.The weighted coefficient is related to the temporal distortion propagation factor of the current CTU.Then,an accurate CTU-level R-D model estimation method was proposed.Finally,an optimal CTU-level bit allocation equation combining the R-D model and temporal distortion propagation factor was established.Its optimization goal is to minimize the distortion of the whole video sequence rather than the current frame under the constraint of frame level bit budget.In addition,in the process of encoding a frame,a relaxed constraint algorithm(RCA)and a strict constraint algorithm(SCA)are employed to reduce the difference between numbers of actual and target bits for the current frame.The SCA achieves very high rate control accuracy,and obtains R-D gains of 3.7% and 4.2% compared with the default HEVC encoder without rate control.The rate control accuracy of the RCA is slightly lower than that of the SCA,but the RCA obtains R-D gains of 5.1% and 5.4% under the configuration of low-delay B and P frames,respectively.
Keywords/Search Tags:high efficiency video coding(HEVC), rate distortion optimization(RDO), rate control, rate distortion dependency, rate distortion model
PDF Full Text Request
Related items