Font Size: a A A

Research On Rate-distortion Optimization Techniques In AV1 Video Coding Standard

Posted on:2022-07-18Degree:MasterType:Thesis
Country:ChinaCandidate:H QinFull Text:PDF
GTID:2518306524476364Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
AV1 is a new-generation video coding standard launched by the open source media alliance AOM in 2018.It aims to develop an open source,patent-free video codec applied to the Internet.AV1 is the first video coding standard of the Open Source Media Alliance.Compared with the previous generation video coding standard H.265/HEVC and Google's VP9,AV1 has improved coding performance under the same high-definition video sequence and test conditions.This thesis mainly studies and analyzes the rate-distortion optimization technology of the AV1 video coding standard,and improves the coding gain based on different models and quality evaluation methods.The thesis consists following research work:(1)For the hierarchical structure of AV1,from the perspective of distortion changes,first list the temporal rate-distortion optimization problems between different temporal layers,and then estimate the aggregate distortion of different temporal layers,and then adjust the Lagrangian multiplier according to the distortion and complete the temporal rate-distortion optimization.At the optimization level of I frame and ARF(Alternative Reference Frame),the reference utilization rate is calculated,and the adaptive optimization of I frame block-level QP is proposed.As AV1 uses two-pass encoding by default,according to the inter-frame information recorded during the first encoding adjusts the model parameters adaptively.Compared with AV1,the proposed method can achieve 2.00% BD-Rate(PSNR)saving in average in the luminance component.(2)Aiming at the hierarchical structure of AV1,starting from the point of view of code rate changes,first establish a model of distortion and code rate change,and then integrate the model into the rate-distortion cost function,and finally adjust the Lagrangian multiplier to complete the temporal rate-distortion optimization.In the analysis and calculation of the original motion compensation error,the reference frames corresponding to different time domain levels are considered,and adaptive optimization is performed.At the same time,the threshold required for the calculation is adjusted according to the inter-frame information recorded in the first encoding,and finally the coding efficiency is improved.Compared with AV1,the proposed method can achieve1.50% BD-Rate(PSNR)saving in average in the luminance component.(3)Aiming at the rate-distortion optimization problem of AV1 in subjective quality,combined with the JND model,analyze and improve the traditional JND model from the perspective of temporal and spatial,and integrate the temporal scaling factor into the model to better reflect the temporal feature.In terms of motion search,the subjective distortion is obtained by using the motion estimation based on the JND threshold,and the optimization of multiple reference frames is added according to the AV1 coding structure,and then the Lagrangian multiplier based on the JND model is adjusted to achieve the rate under the subjective quality distortion optimization.Compared with AV1,the proposed method can achieve 1.18% BD-Rate(SSIM)saving in average in the luminance component.
Keywords/Search Tags:AV1, High Efficiency Video Coding(HEVC), video coding, rate-distortion optimization, JND model
PDF Full Text Request
Related items