TY  - JOUR
AU  - Shi, Kehan
TI  - Adaptive total variational regularization of Gaussian denoisers for multiplicative noise removal
JO  - Computers and mathematics with applications
VL  - 168
SN  - 0898-1221
CY  - Amsterdam [u.a.]
PB  - Elsevier Science
M1  - PUBDB-2024-07527
SP  - 207-217
PY  - 2024
AB  - In this paper, a convex variational model based on the adaptive total variation (TV) regularization is proposedfor image restoration under multiplicative noise. The adaptive weight allows for greater smoothing in the brightregion for the suppression of speckles. The model includes a nonconvex data fidelity term and also a quadraticpenalty term that enforces the restored image to be close to a reference image deduced from a Gaussian denoiser.It can be viewed as the adaptive TV regularization of the Regularization by Denoising (RED) approach formultiplicative noise removal. We prove that the model admits a unique minimizer in a suitable function spaceand provide a fast numerical algorithm based on the alternating direction method with multipliers (ADMM) forit. Different Gaussian denoisers, including the patch-based algorithm BM3D and the learning-based algorithmDnCNN, are considered for the model in numerical experiments. It is shown that our model efficiently removesmultiplicative noise without introducing artifacts.
LB  - PUB:(DE-HGF)16
UR  - <Go to ISI:>//WOS:001271428200001
DO  - DOI:10.1016/j.camwa.2024.06.017
UR  - https://bib-pubdb1.desy.de/record/619279
ER  -