Unpaired Multiview Video plus Depth compression

Abstract

Recent developments of three-dimensional (3D) video coding greatly rely on the use of Multiview Video plus Depth (MVD) data format for representing a 3D scene. This type of data can be coded with conventional video compression schemes and can enable advanced 3D video functionality, such as rendering of virtual views at the decoder side. The MVD represents a 3D scene from different viewing angles as video and pixel-wise associated depth data. In this paper we consider the redundancy of depth data in the MVD representation and propose a novel scheme, called Unpaired MVD (UP-MVD) format to be used in 3D video applications. Being a subset of the MVD this new format assumes that a reduced number of depth views compared to the number of texture views can reduce bitrate as well as the encoding/decoding complexity while still providing the 3DV functionality in many scenarios. The simulation results show that the proposed unpaired MVD format outperforms the MVD format on average from 0.9% to 6.95% of Bjontegaard delta bitrate (dBR) for the baseline disparity adjustments from 50% to 100% of the coded baseline, respectively. Moreover, UP-MVD provides equal or better rate-distortion results for all test sequences for up to 20% view separation adjustment, and in five out of seven sequences a better rate-distortion performance is observed when 50% view separation adjustment is applied.