| Efficient video coding is a key basic technology for many digital media applications.The new generation HEVC(High Efficiency Video Coding)video coding standard uses multiple efficient coding technologies and shows excellent coding compression performance,includes a coding unit(CU)based on a quadtree,transform unit(TU),And according to different prediction modes,a prediction unit(PU)with a more flexible division method.In the case of the same video coding quality,the compression efficiency is doubled than the previous generation video coding standard H.264.In actual applications,due to the limitations of transmission bandwidth and encoding complexity,the Rate Distortion Optimization(RDO)criterion is commonly used in video encoders to optimize the design of encoding algorithms.In HEVC,the coding unit size can range from 8x8 to64x64,which is more flexible than the H.264 fixed-size 16x16 macroblock.This coding structure effectively reduces the bit rate overhead,especially when encoding high-resolution video.The CU/PU/TU combination scheme can greatly improve the coding efficiency,but RDO-based mode decisions need to traverse various combination schemes,and perform bit rate and distortion cost calculations for a large number of candidate modes,which involve forward transformation,quantization,inverse quantization,and inverse transformation,which leads to high computational complexity and increased hardware overhead.In some application scenarios that require real-time encoding,this encoding method is not applicable.Using rate model and distortion model can simplify the calculation of rate-distortion cost,save hardware resources,and better adapt to real-time coding scenarios.Therefore,it is very important to construct a high-accuracy rate and distortion model.However,because the HEVC entropy coding algorithm uses CABAC,causinga serious dependence between the coefficients.With the update of the probability state of the context model,even the same coefficients will cause different rate consumption.This feature makes it difficult to make a systematic analysis of the code rate.At the same time,because each syntax element uses different context models and encoding methods,it is more difficult to construct a high-accuracy rate model.When calculating the distortion,a lengthy reconstruction process is involved.Therefore,how to provide a simple reconstruction algorithm with a smaller error in the pre-vectorization process and estimate the distortion is also a major problem in the current academic community.Facing the current situation that the calculation cost of rate and distortion in HEVC is too high to apply to the hardware level,there is a need for the research of rate and distortion estimation algorithms.The research in this article focuses on solving these two problems: one is to analyze the dependence of the context model on the rate calculation from the principle,quantitatively describe this dependence,and establish the rate prediction model to solve the dependency problem between coefficients.The other is to optimize the reconstruction process in the distortion calculation,establish a distortion model according to the transform domain reconstruction parameters,and use the rate and distortion model in the RDO process to simplify the calculation of the bit rate and distortion in the RDO process.The main research work of this paper includes:1.Propose a rate model: Explore the coding principle of CABAC and the root cause of its dependence and studying.Focus on the analysis of syntax elements with a relatively large rate proportion,and then propose location parameters to describe the rate difference between the high and low frequency coefficients.Under different size transform units,different characteristic parameters are used,and the transform block-level code rate model is constructed through offline mathematical modeling.For the code rate of the header information,the context probability model selection condition is used as the classification basis,At the same time,the method of fixed context probability state is adopted for individual information,and the relationship with the quantization parameter(Quantization Parameter,QP)is explored,a header information rate model is proposed to provide model support for the code rate calculation in the RDO process.2.Propose distortion model and RD estimation algorithm: study the distortion algorithm in HEVC,use transform domain parameters for simple reconstruction,instead of calculation in space domain to save the calculation process of inverse discrete transform and image reconstruction,analyze the relationship between distortion and quantization parameters,and adopt adaptive weight coefficients to change with the change of QP to enhance the accuracy of the model.The code rate and distortion model is applied to the RDO process to reduce its computational complexity.Experimental results show that the proposed rate and distortion estimation algorithm can achieve a complexity reduction of about 50%,while BD-BR only increases by1.74%... |
PDF Full Text Request