Overlapping group prior for image deconvolution using patch-wise gradient statistics

Open Webpage

Haegeun Lee, Hansol Kim, Jinook Lee, Moon Gi Kang

Published: 01 Jan 2023, Last Modified: 10 Nov 2025Signal Process. 2023EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•The proposed overlapping group prior considers textures and details of an image based on patchwise gradient statistics in non-blind image deconvolution problem.•The developed prior update strategy naturally estimates the innate regional information without any external dataset or additional trials.•The spatially variant and non-convex objective function is efficiently minimized in the ADMMframework, transforming the patch-wise notations to the pixel-wise representations.•The proposed method demonstrate higher performance in restoring the high frequency components, such as textures, patterns, and details, compared to the conventional algorithms.