Quality analysis of virtual views on stereoscopic video content

Depth Image-Based Rendering (DIBR) allows the creation of virtual camera viewpoints from a 2D image and its corresponding disparity map. This enables a variety of new applications in 3D film post-production where disparities can be computed in an automatic way from the stereoscopic content. Examples include scene depth correction, content re-mastering and multi-view generation for auto-stereoscopic displays. In this paper, a comparison of state-of-the-art DIBR techniques in the context of 3D video adaptation is presented. We first provide an evaluation method that enables subjective comparison of the visual quality of DIBR-generated 3D film sequences. Based on this, we then evaluate the impact of image artifacts on the visual comfort and the depth impression for four different DIBR approaches.