Evaluating multi-view plus depth coding solutions for 3D video scenarios

3D and free viewpoint video (FVV) are new media types able to provide the user an experience far beyond what is currently offered by traditional 2D video systems. Among the possible formats, the so-called multi-view plus depth format has the potential to synthesize at the decoder more views than those explicitly coded at the encoder. These additional views can be used to render arbitrary viewpoints (for FVV) or used in auto-stereoscopic displays, where a partial set of views are synthesized. Since depth provides information on the 3D scene geometry, depth coding artifacts (errors) affect the performance of the view synthesis techniques and may lead to geometric distortions in the virtual intermediate views. This paper makes a comprehensive performance evaluation of the relevant 2D texture coding schemes applied to both the texture and depth data, providing a much needed reference study. With this purpose in mind, two scenarios have been defined depending on the allowed encoder complexity. For each of these scenarios, the most relevant 2D video codecs have been evaluated, notably the low encoder complexity DISCOVER distributed video codec and the latest High Efficiency Video Coding (HEVC) standard predictive video codec. The performance of these coding solutions is evaluated through the quality of the rendered view according to precise and widely used test conditions.