Fast Depth Map Intra Coding Based Structure Tensor Data Analysis

Abstract

As a recent 3D video coding standard, ISO/IEC MPEG and ITU-T Video Coding Experts Group (VCEG) establish 3D-HEVC as the most efficient 3D video coding based on Multiview texture Videos plus Depth maps data format. In 3D-HEVC, depth map intra prediction is a key factor in 3D video coding, in which, the encoder utilizes the conventional intra prediction and depth modeling modes together to improve the depth map coding. This improvement of depth map intra prediction increase the coding efficiency significantly, but result in a dramatic computational complexity load, due to the exhaustive searching for the best intra mode. The increase of the intra coding complexity excludes the 3D-HEVC from real time and real world application. To resolve the aforementioned problem, it's imperative to develop solutions that can reduce the complexity meaningfully. In this work, we propose an efficient depth map intra prediction model decision based on tensor features. The simulation experiments prove that the developed model decreases the computational complexity (38.52%) with no performance losses.