The next generation 3D video systems employ the view synthesis method that generates virtual viewpoint images using depth data to provide realistic and comfortable depth impression. It is suitable for the advanced stereoscopic displays or the auto-stereoscopic displays since it can generate multi-view images. To provide high-quality virtual view images, the 3D video system should minimize the visual artifacts induced by the hole regions. Particularly, the visual artifacts arise severely during view extrapolation due to lack of information from the reference views. In this paper, we propose an efficient extra-view generation method, which is useful for the auto-stereoscopic displays in the circumstance that the number of provided viewpoint is not sufficient at the displays. First, we propose a framework that generates outer viewpoint images. Second, we propose an efficient hole filling methods in spatial and temporal directions. Using objective and subjective evaluations, we showed that the proposed method generate high-quality extraviewpoint images.
展开▼