VMAF Orientated Rate-distortion Optimization For HEVC

The tremendous advances in computer and communication technology have brought about a proliferation of digital media content.Video compression is still an important technology to alleviate content and bandwidth conflicts.High Efficiency Video Coding(HEVC),compared with the previous generation standard,has greatly improved the overall coding performance and has been widely used.Therefore,it is of great practical value to optimize the HEVC encoder.At present,in the rate-distortion optimization of HEVC,the traditional objective video quality evaluation index is often used to guide the coding decision-making process,and the visual characteristics of the human eye are not considered.As a result,the final optimization result may have a large error with the subjective perception quality of the human eye.Aiming at this problem,this thesis studies the rate-distortion optimization method based on the subjective perceptual quality evaluation index VMAF(Video Multi-Method Assessment Fusion),and proposes two solutions to introduce VMAF into the rate-distortion decision-making process.The specific research contents and contributions are as follows:(1)VMAF orientated rate-distortion optimization based on postprocessing information: Considering the correlation between SSE and VMAF distortion,this thesis proposes to use video post-processing information to correlate SSE and VMAF distortion,and adjust Lagrange multiplier through frame-level and block-level to construct VMAF-based methods.A rate-distortion optimization framework for distortion metrics to improve encoding compression performance.Experimental results show that this method saves 2.72%-3.11% of the average bitrate under the same subjective VMAF quality at the cost of 211%-212% of the original encoding time.(2)VMAF orientated rate-distortion optimization based on the just noticeable difference: In addition,considering the high time complexity of existing methods,starting from the human visual characteristics behind the VMAF evaluation standard,this thesis improves the just noticeable difference model by synthesizing various masking effects of human visual perception,and proposes The method of adjusting the Lagrange multiplier by correlating the VMAF distortion with the just noticeable difference can build a rate-distortion optimization framework based on the VMAF distortion metric to improve the coding and compression performance.Experimental results show that under the same subjective VMAF quality,the method saves 2.91% of the average bitrate,and the encoding time is only 105% of the original.All in all,the two methods proposed in this thesis can well introduce VMAF perceptual distortion into the HEVC rate-distortion optimization process,and solve the inaccurate quality measurement problem in traditional rate-distortion optimization.Bit rate saving is achieved under the same subjective VMAF quality,which has good application value in relieving network bandwidth pressure.