Rhombic Mapping Scheme for Panoramic Video Encoding

Abstract

Since modern video coding standards are not designed for processing panoramic videos, the spherical videos are mapped onto a rectangular plane before encoding. Usually the mapping scheme includes two steps: sampling points on the sphere and arranging the points into a rectangle. Traditional mapping schemes, such as equirectangular and cubic mapping, have different sampling densities on the sphere, which creates many non-information-carrying pixels. Besides, the cubic mapping leads to some discontinuous (adjacent but uncorrelated) pixels, which are not conducive to inter and intra prediction of the encoding, when arranging the six cube faces into one rectangle before encoding. In this paper, we propose a novel rhombic mapping scheme that not only solves the issue of oversampling, but also arranges the points into a compression-friendly rectangle without discontinuous pixels. Experimental results demonstrate that the proposed mapping scheme can save more bitrate compared to the state of the art mapping schemes.