An efficient spherical video sampling scheme based on Cube model

Abstract

Virtual reality (VR) is rapidly arising in various fields, such as navigation, robotics, and documentation. Spherical panoramic video, instead of 3-dimensional (3D) modeling, provides immersive and omnidirectional views in a much more convenient way. However, state-of-the-art video encoding techniques such as HEVC cannot be applied directly to spherical videos. In traditional ways, spherical videos are often projected onto a plane or a cube for the convenience of encoding, but the mapping quality and encoding efficiency are not considered. In this paper, a Cube-Snyder (CS) projection is proposed, along with a gradually-varied sampling method, to generate panoramic video, with which visual quality is improved and a more uniform pixel distribution is obtained. By applying Cube-Snyder projection with gradually-varied sampling (GVCS) method, average PSNR is raised by 0.2 to 0.5, compared with traditional mapping schemes.