Research On Virtual View Generation For Natural 3DTV Systems

As a novel digital media of stereoscopic vision, natural three-dimensional television (Natural 3DTV) expands the viewer experience beyond what is provided by traditional media and has been attracting wide attention of research communities. Virtual view generation is a key technology for future Natural 3DTV applications. It makes viewers to perceive the depth effect of the observed scenery, while allowing them to change their viewpoints interactively.This paper focuses on the research of virtual view generation for Natural 3DTV systems. In order to generate virtual views with high image quality, the paper utilizes the technology of depth-image-based rendering (DIBR), which is considered having much potential for view rendering. The main achievements of the paper are listed as follows:(1) A new and effective virtual view generation method for Natural 3DTV systems is proposed. Based on the technology of DIBR and essential occluded information, the method could generate arbitrary virtual views with high image quality as well as low algorithm cost and bandwidth requirement. The method first promotes a distance-depth-weighted (DDW) algorithm to optimize the DIBR algorithm, and then performs an asymmetric bidirectional synthesis process to create virtual view images effectively by solving the problems of visibility, re-sampling, and disocclusion of the basic DIBR process.(2) A new way of representation for occluded information is proposed. It could effectively express the occluded data and could be used to fill the holes during the process of virtual view generation. Based on this representation, a more accurate method is presented by increasing a small amount of depth information.