Adaptive 3D content generation via depth video based rendering

摘要

Depth-Image-Based Rendering (DIBR) is a technique which utilizes 2D texture images and corresponding depth maps to synthesize stereoscopic views. However, it could introduce the hole artifacts that heavily degrades the rendered quality of 3D video. In this paper, a content-adaptive Depth-Video-Based Rendering (DVBR) method is proposed to eliminate the hole artifacts that are caused by quantization and occlusion/disocclusion. For the tiny holes raised by quantization error, we adopt image inpaiting technique. For the large holes due to the occlusion problem in stereopsis, we fill the hole use of the temporal correlation. Experimental results show that the proposed DVBR method improved spatial quality and eliminates the temporal artifacts by the content-adaptive hole-filling schemes, resulting in better visual quality for the rendered 3D video, which leads to better visual quality for the rendered 3D video.