Centre for Mathematics, University of Coimbra

Variational multi-channel inpainting and denoising model: existence of solution and numerical approximation (Preprint)

<Reference List>
Type:	Preprint
National /International:	International
Title:	Variational multi-channel inpainting and denoising model: existence of solution and numerical approximation
Publication Date:	2020-10-06
Authors:	- Isabel Narra Figueiredo - Mahdi Dodangeh
Abstract:	In this paper we propose a variational "total variation like" inpainting and denoising model, for multi-channel images, prove the existence and uniqueness of its solution, and define and implement a numerical scheme for its solution. This variational model includes, besides the data-fidelity term, an extension of the total variation regularizer, appropriate for vector-valued images, aiming at reconstructing sharp image edges, and also a smooth regularizer for removing noise. The proposed numerical algorithm is an instance of the so-called alternating direction method of multipliers, for fast image recovery, which transforms the discrete version of the variational model into a constrained optimization problem, by using variable splitting. This constrained problem is subsequently solved by an augmented Lagrangian approach. For assessing the method, tests are performed on real-world and good quality color images, that are artificially damaged by adding noise and removing randomly pixel values in the different image channels. In addition an experiment on ab initio degraded medical image, corrupted with specular highlights, is also carried out.
Institution:	DMUC 20-28
Online version:	http://www.mat.uc.pt...prints/eng_2020.html
Download:	Not available