| As one of the most important data forms of Internet traffic,video needs video codec technology to compress redundant information in order to be transmitted efficiently.In the process of the development of video coding technology,there have been several codec standards based on hybrid video coding framework,such as H.264,H.265,H.266,etc.The new standard achieves the purpose of improving the performance of encoders by adding new coding tools and techniques.In this thesis,some key techniques of multiple encoders under different standards are optimized.The specific work is as follows:(1)Aiming at the loop filtering technique of H.266 standard,the improved Denoising Convolutional Neural Network is proposed to replace the Sample Adaptive Compensation module and Deblock Filter module.Then 256×256 blocks are used by block filtering and effectively reduce the distortion caused by block effect and ringing effect in the encoding process.The method is applied to VTM18.0 test platform of H.266,and the average BD-Rate coding gain of 2.84% for Y components is obtained by using standard video sequences.(2)Aiming at the rate control technique in x264 encoder,a zero-delay rate control method based on R-λ model and time-domain dependency is proposed.Firstly,in the aspect of frame-level bit-rate control,dynamic bit allocation is carried out for the coded frame using a sliding window,and then the frame-level Lagrange multiplier and framelevel quantization parameters are obtained by using the fitted R-λ model and QP-λ curve to achieve the target bit.Finally,the parameters of the model are updated.At the same time,a reverse time-domain distortion propagation chain is established at macroblock level,and the Lagrange multiplier is adjusted by using the obtained back propagation factor,thus realizing the rate-distortion optimization at the macroblock level.Compared with the fixed QP mode of x264,this method brings 0.84% BD-Rate coding gain to the luma component.Compared with the ABR(Average Bit Rate)mode of x264,this method reduces the bit error rate to 0.07% and reduces the complexity by 7.26%.(3)Aiming at the motion estimation technique in x265 encoder,a quarter-pixel motion search optimization method based on quadratic error surface is proposed.First,a six-parameter quadric error surface model is established,then six parameters of the model are obtained by using the rate distortion cost of the optimal integral pixel and eight adjacent integral pixel points.In this way,we can get the search center of the quarter pixel based on the error surface equation and six parameters,and the optimal sub-pixel point(the search center of quarter-pixel motion search)is obtained.The method is applied to the ultrafast gear of the x265 encoder,and the average BD-Rate encoding gain are 9.86%for the Y components using standard video sequences. |
