Review and Preview: Disocclusion by Inpainting for Image-Based Rendering

摘要: Image-based rendering takes as input multiple images of an object and generates photorealistic from novel viewpoints. This approach avoids explicitly modeling scenes by replacing the phase with reconstruction phase. Reconstruction is achieved in two possible ways: recovering 3-D point locations using multiview stereo techniques, or reasoning about consistency each voxel a discretized volume space. The most challenging problem for image-based presence occlusions. Occlusions make ambiguous parts not visible any image. These must be reconstructed visually acceptable way. paper both reviews image inpainting argues that can provide only attractive but also framework increasing accuracy depth recovery. Digital refers to methods fill-in holes arbitrary topology so they seem part original Available are broadly classified structural textural inpainting. Structural reconstructs prior assumptions boundary conditions, while considers available data texture exemplars other templates. Of particular interest research on applied models, emphasizing its effectiveness disocclusion.