Research On Optimization For Panoramic Video Coding

Panoramic video,also known as omni-directional video or 360° video,features a high frame rate and high resolution.The video content can be mapped to a sphere through a head mounted display,and the audience can have a 360 ° immersive visual experience.It is precisely because of the unique characteristics of panoramic video that more bit overhead is required in the video transmission,which lead to a heavy burden on storage and bandwidth.Most of the existing panoramic video coding optimizations did not take into account the effect of pixel position on the Lagrange multiplier in the rate-distortion optimization(RDO).Meanwhile,most of the existing methods do not use the temporal influence factor to regulate the rate distortion optimisation process.This thesis optimizes and improves the HM encoder in the following aspects for panoramic video:(1)For the rate-distortion optimization of video coding,we proposed a latitudeguided temporally dependent RDO algorithm for panoramic video.Firstly,the quantization parameter(QP)of the coding unit(CU)is determined by the latitude corresponding to the pixel.Secondly,the Lagrange multiplier of the corresponding coding tree unit(CTU)is calculated by the QP of the CTU.Finally,the Lagrange multiplier is fine-tuned to improve the RDO process by using the temporal dependency among CTUs.Experimental results show that,on the platform of H.265 test model(HM-16.14),the proposed algorithm obtains bitrate savings of 4.15%,4.14% and 4.05% on average under the quality assessment criteria of Spherical PSNR(S-PSNR),Weighted to Spherically Uniform PSNR(WS-PSNR)and Craster Parabolic Projection PSNR(CPP-PSNR)for the equirectangular projection(ERP).(2)For the rate control process of video coding,we propose a spherical coordinate and temporally dependent algorithm for controlling the bit rate of panoramic video.the ERP projection format is characterized by increasing pixel redundancy as the latitude increases.In this thesis,we consider the impact of its pixel distribution on the coding effect,and take the pixel position information in the sphere domain into account in the CU block division of the coding process,making the block division more refined.The influence of latitude distribution on the CTU-level quantization parameter QP is also taken into account in the rate control process,and latitude information is added as a weight in the QP update process.Finally the influence of the temporal impact factor on the CTU-level Lagrangian multipliers is also considered.The algorithm proposed in this work is implemented on the highly efficient video coding H.265 reference software HM-16.20 and achieves average code rate savings of 14.61%,14.63% and 14.89% in the S-PSNR,WS-PSNR and CPP-PSNR quality metrics,respectively,in the low delay P-frame(LDP)configuration compared to the benchmark encoder.