Depth inpainting scheme based on edge guided non local means
Adhi Prahara, Andri Pranolo · 2017
The acquisition process of depth image usually produces noise and holes. This kind of defect can reduce the quality of depth image especially the existence of small or large holes which cause great loss of depth information. In this research, the depth inpainting scheme based on edge guided non local means is proposed to restore the missing depth information. Non local means (NL-means) uses similar patches within search window to recover the missing depth pixel by averaging the weighted Euclidean distance from the similar patches. The inpainting scheme iterates through boundaries of missing depth pixels. Depth image edges will act as guidance to limit the search region by using Breadth First Search (BFS). The distance calculation is performed on valid depth pixels and ignore the missing depth parts from both patches to avoid false depth information. The experiment on the depth image dataset shows that the proposed scheme can fill small to large holes on depth image with low MSE value.