Interframe hole filling for DIBR in 3D videos

Abstract

Depth-image-based rendering (DIBR) produces multiple views efficiently. However, its process lacks some viewpoint information. There will be holes, which influences the 3D video quality. Previous DIBR techniques were mainly applied to 3D images, so it only relied on the single view and the depth map to fill the holes, but insufficient repair information resulted in incorrect repair. In this paper, we aim to repair 3D videos by the hole filling for DIBR, which introduces the relation between frames to increase the repair information. We apply the moving behavior and the texture similarity within interframes to assure the accuracy of the repair information. The experiment results demonstrate that compared to previous methods, the proposed method obtain better 3D video quality. The average peak signal-to-noise ratio (PSNR) increases by 1.014 dB, and the structural similarity (SSIM) index increases by 0.012, which shows that the proposed method obtains better quality than the methods that only apply the single view image information.