Artifact Removal Algorithm for Virtual Viewpoint Images Based on Occlusion-Aware Cavity Dilation

Free Viewpoint Television (FTV) enables viewers to select any viewing angle, enhancing interaction and immersion. Depth Image Based Rendering (DIBR) is notable for reducing the number of transmitted viewpoint images and quickly generating virtual views. However, artifacts in these images degrade quality significantly. This paper proposes an artifact removal algorithm for virtual viewpoint images based on occlusion-aware cavity dilation, addressing the shortcomings of existing methods that indiscriminately apply morphological expansion to all cavities, leading to unnecessary pixel loss. Instead, our proposed method classifies cavity types, selectively applies morphological expansion to the occlusion cavities, and promptly restores the covered foreground edge pixels with their original pixel values. Experimental results show that this algorithm can effectively handle artifact issues in virtual viewpoint rendering, achieving higher quality virtual viewpoint views and outperforming other algorithms in terms of PSNR and SSIM.