Robust depth map inpainting using superpixels and non-local Gauss–Markov random field prior
作者：
Highlights：
• Inpainting of depth maps by exploiting a non-local extension of Gauss-MRF prior.
• Reasonable reconstruction even when 90% of the depth data is randomly missing.
• Handle large occlusions and avoid smoothness particularly at depth edges.
• Depth map inpainting guided by multi-scale superpixel of respective RGB frame.
• Non-local superpixel based search window, insted of traditional rectangular window.
摘要
•Inpainting of depth maps by exploiting a non-local extension of Gauss-MRF prior.•Reasonable reconstruction even when 90% of the depth data is randomly missing.•Handle large occlusions and avoid smoothness particularly at depth edges.•Depth map inpainting guided by multi-scale superpixel of respective RGB frame.•Non-local superpixel based search window, insted of traditional rectangular window.
论文关键词：Depth map inpainting,RGB-D camera,Superpixel,Markov random fields
论文评审过程：Received 29 August 2020, Revised 24 April 2021, Accepted 6 July 2021, Available online 13 July 2021, Version of Record 19 July 2021.
论文官网地址：https://doi.org/10.1016/j.image.2021.116378
