An Approach to Distributed Video Coding Using 3D Face Models

Distributed source coding (DSC), formulated thirty years ago, is lately witnessing a great research effort. This effort is encouraged by emerging new applications that could greatly benefit from the properties inherent to DSC, like low complexity encoders and embedded error resilience. Among the challenging topics related to DSC there is the generation of the side information, an estimation made by the decoder of the information being decoded. This paper quickly summarizes the theoretical bases of DSC and then focuses on the essential task of generating the side information, by developing the model-based method drafted by the authors on a previous publication. It is shown that the application of model-based side information generation techniques on top of current state of the art DSC-based video codecs provides an improvement of 0.5 to 0.75 dB over a wide range of bit-rates. Comparisons against H.264 are also provided.