Coding Techniques for Multi-View Video Signals

Multi-view video is a technique where a moving object or a scene is recorded using a setup of several synchronous cameras. By applying image-based rendering techniques, new views can be generated for arbitrary viewpoints and viewing angles. Even complex lighting conditions and surface characteristics are reproduced in a photorealistic quality. Furthermore, the computational effort is independent from the scene complexity. Applications include three-dimensional television, where the viewer can choose his viewing position and viewing angle freely, as well as medical visualisations. However, recording multi-view video signals leads to massive amounts of data. That is why efficient compression is crucial for making practical systems possible.

The aim of this thesis is on the one hand the efficient coding of multi-view video data and on the other hand the development of pre- and post-filtering techniques which improve the compression and the rendering result. Instead of coding each camera view separately, the correlation between neighbouring views is exploited. The achievable gain by doing so is estimated through a statistical analysis. Furthermore, the standardisation efforts of the Joint Video Team (JVT) regarding multi-view coding are taken into account. Further aspects which are investigated include random access to the coded data and the compression of depth maps, which are frequently used as side information.