Intra prediction based on geometry padding for omnidirectional video coding

Omnidirectional videos or 360 degree videos play an important role in virtual reality (VR) applications. In order to employ the existing video coding standards, omnidirectional videos are firstly projected onto the 2-Dimension (2D) plane, which generates the discontinuity at boundaries and may result in unexpected artifacts when lossy coding is applied. In this paper, we propose a new intra prediction method to deal with the above coding artefacts in omnidirectional videos. Different from the conventional intra prediction using the left and top reference samples, the right reference samples are derived from the reconstructed samples on the left boundary and padded for intra prediction. The proposed method applies to the planar, DC and partial angular prediction modes in intra prediction. The experimental results demonstrate a Bjøntegaard-Delta-rate reduction of up to 2.97% using weighted spherical peak-signal to noise ratio (WSPSNR) quality metric for the coding tree units (CTUs) at the right boundary.