A fast inter-component depth modeling technique for 3D high effiency video coding

Abstract

Depth maps are made up of sharp edges and near homogenous areas. To better represent object edges in depth videos, the three-dimensional (3D) extension of the High Efficiency Video Coding (HEVC) standard allows the use of additional Depth Modeling Modes (DMM) to the existing intra-prediction modes. One of these modes, Mode 3, performs distortion calculations on a set of pre-defined Wedgelet lists and signals the index of the Wedgelet that presents the least distortion. This paper proposes a fast and less complex Mode 3 depth-modeling scheme for the 3DHEVC encoder, where the set of Wedgelets is reduced to just six. This is achieved by acquiring four high-intensity changing points from the reference block. A 4.7% and 1.7% reduction in encoding runtime is achieved when testing this method with the hierarchical and all intra-prediction structures, respectively. In both cases, the resulting Peak-Signal-to-Noise Ratio (PSNR) values are maintained.