Research On Block Partition Technology In 360-degree Video Coding

In recent years,with the growth of computer three-dimensional processing capabilities and the emergence of low-cost sensor display components,virtual reality has developed rapidly.As commercial interests in deploying VR applications rise,360-degree video has been attracting increasing attention as a new type of media that provide immersive experience,becoming a current hot topic in computer science and industrial applications.With the development of virtual reality technology,360-degree video coding has gradually become a popular research direction.In JVET which is exploring next generation of video coding standard with capability beyond HEVC,360-degree video is included in the scope of standard along with SDR and HDR video.Comparing with traditional video,360-degree video has higher resolution and larger data amount,which brings a huge challenge to practical applications.The problem of coding efficiency remains to be solved.This paper proposes two optimization algorithms for block partition in 360-degree video coding.The first is a 360-degree video fast encoding algorithm based on the coding tree unit latitude.When the 360-degree video is projected to Equi-rectangular format,it may be stretched and the sampling rate of two-pole is high but the equator is relatively low,thus our algorithm divides the ERP format into three regions,and determines the depth range of each region according to the sampling rate.When the algorithm is executed,the latitude of the current CTU is firstly calculated according to the latitude of each pixel,then the CTU is mapped to one of the defined region according to the latitude to obtain the depth range value.Compared with the anchor HM16.15-360 Lib,the algorithm can achieve 11.9% encoding time reduction under the contition of RA10 configuration and end-to-end WS-PSNR evaluation criteria,with an average increase of 0.6% in luminance component BD-rate.The second is a fast depth decision algorithm based on the adjacent CTU depth and the current CTU gradient.The algorithm consists of two parts.The first part is the prediction process of the current CTU depth range based on the adjacent CTU depth.The proposed algorithm uses the maximum and minimum depth information of the adjacent CTUs.Taking into account the high resolution characteristics of the 360-degree video,the algorithm further narrows the depth range.The second part is to adjust the depth range predicted by the first part according to the gradient of the current CTU,mainly for the CTU with narrow depth range and simple texture CTU.The two parts are combined with each other to obtain the final depth range of the CTU.With the RA10 configuration and the end-to-end WS-PSNR evaluation criteria,this algorithm can achieve a 29.1% coding time reduction when the luminance component has only 0.3% of encoding performance loss.