Three-dimensional natural video system based on layered representation of depth maps

Abstract

In this paper, we propose a new three-dimensional (3-D) video system using a depth image-based representation (DIBR). In order to represent 3-D natural videos, we exploit texture images and synchronized depth maps, which are decomposed into four layers; number of layer (NOL), grid, boundary and feature point layers. With NOL, grid, and boundary layers, we define minimum required consecutive surfaces for 3-D natural videos. Then, we enhance their visual qualities by adding feature point layers. We regard NOL, grid and boundary layers as a base layer and feature point layers as enhancement layers to construct multi-layer representation of 3-D natural videos. With our proposed system, we cannot only represent depth maps efficiently, but also render 3-D natural videos in real time within reliable visual qualities. In addition, 3-D natural videos can be reconstructed adaptively according to consumers' capabilities and target applications within the framework of multi-layer representation. Experimental results have demonstrated that our proposed system can render and compress 3-D natural videos efficiently, as well as supporting the functionality of multi-view image generation