Disocclusion

Disocclusion refers to the appearance of parts of a scene that were previously hidden behind another object, when the viewpoint changes or objects move. It is the complement of occlusion. In computer vision and graphics, disocclusion concerns arise in stereo depth estimation, multi-view reconstruction, and view synthesis, where pixels visible in one view may be invisible in another due to foreground occluders. The resulting disoccluded regions in disparity maps or synthesized views represent areas where information is missing and must be inferred to produce coherent reconstructions or renderings.

In stereo matching and multi-view tasks, disoccluded areas lack correspondences in the counter view and cannot be directly matched. They are often filled using inpainting, texture synthesis, or by leveraging temporal or multi-view cues. Approaches to handle disocclusion include explicit modeling of visibility with occlusion or fronto-parallel maps, left-right consistency checks, and using surrounding regions, smoothness priors, or learned priors from data to fill gaps.

Challenges in disocclusion handling include large disoccluded regions, textureless or repetitive areas, depth-ambiguity, and ensuring photometric and structural consistency across views or frames. Evaluation is also difficult because ground truth for occluded content may be unavailable or uncertain. Applications span 3D reconstruction, virtual view synthesis, augmented reality, and image-based rendering, where accurate treatment of disocclusion improves realism and stability of the produced views.