Perceptual quality driven 3-D video over networks
3-D video in day to day life will enhance the way we represent real-world sceneries and provide more natural conditions for human interaction. Therefore, 3-D video has the potential to be the next killer application in multimedia communications. However, the demand for resources (e.g. bandwidth), 3-D quality evaluations and providing error protection are challenges to be addressed. Thus, this thesis addresses the issues related to transmission of 3-D video over communication networks including compression, quality evaluations, error resilience and error concealment. The first part of the thesis investigates encoding approaches for 3-D video in terms of compression efficiency and adaptability to existing communication technologies. Moreover, an encoding configuration is proposed for colour plus depth video coding based on scalable video coding principals. The proposed encoding configuration shows improved compression efficiency and scalability which can be utilized to scale conventional video applications into stereoscopic video with a minimum increase to the bandwidth required. Quality evaluation issues of stereoscopic video are addressed in the second part of the thesis. The correlations between objective and subjective quality ratings are derived for the range of compression ratios and packet loss rates considered. The results show high correlation between candidate objective measures (e.g. PSNR of colour image) and the measured 3-D perceptual quality attributes. The third part of the thesis investigates efficient error resilience and concealment methods for backward compatible stereoscopic video transmission over wired/wireless networks. In order to provide enhanced error recovery, the proposed methods utilize inherent characteristics of colour plus depth video and their contributions towards improved perceived quality. The error resilience methods proposed improve 3-D perception compared to equally protected transmission of colour plus depth map video. Similarly, the proposed error concealment methods recover missing information more effectively compared to the deployment of existing 2-D error concealment methods.