Improving perceptual quality of 3D TV systems

3D video has the capability to enhance the multimedia experience of viewers by providing an added sensation of depth. While, stereoscopic 3D video has already made its mark in digital cinema, the entertainment industry is now focused on delivering 3D media to the home and to mobile devices. In the light of these developments, this thesis presents several models and techniques to perceptually optimize production, compression and transmission stages of the 3D video distribution chain. Mathematical models are derived in this thesis to quantify the Just Noticeable Difference in Depth (JNDD) perceived by the human visual system for depth cues provided by stereoscopic displays, such as binocular disparity, retinal blur and relative size. The JNDD models are used to develop a pre-processing technique to reduce production artefacts in depth maps that are used in the virtual view generation process in Depth Image Based Rendering (DIBR) applications. The depth maps used in DIBR applications are represented as gray scale images, but they are not viewed by end users. Therefore, depth maps need to be compressed in a way that is suitable for the virtual view generation process. To this end, this thesis presents three novel techniques to compress depth maps. The first is an object based encoding technique, which exploits the correlations between the color image and its corresponding depth maps. The second technique employs a renderer in the encoder to quantify the rendering distortions and the other is based on an analytical model that approximates rendering errors caused by errors in the depth map. Furthermore, a decoder based post-processing filter is proposed to improve the quality of rendered views by reducing the effects of quantization during depth map compression. Finally, in an experimental investigation of transmission of 3D video, the error propagation characteristics of interview predicted stereoscopic video and the effect of packet losses on DIBR systems is analyzed in this thesis. Significant improvements in perceptual quality and rendering performance have been achieved by the proposed techniques at reduced bit rates required to encode 3D content. It is expected that the proposed techniques will have important use cases in advanced 3D video distribution systems.