Locally adaptive total variation for removing mixed Gaussian–impulse noise

ABSTRACT

The minimization of a functional consisting of a combined $L^1$/$L^2$ data fidelity term and a total variation regularization term with a locally varying regularization parameter for the removal of mixed Gaussian–impulse noise is considered. Based on a related locally constrained optimization problem, algorithms for automatically selecting the spatially varying parameter are presented. Numerical experiments for image denoising are shown, which demonstrate that the locally varying parameter selection algorithms are able to generate solutions which are of higher restoration quality than solutions obtained with scalar parameters.

KEYWORDS:

• Locally dependent regularization parameter
• automated parameter selection
• mixed Gaussian–impulse noise
• combined $L^1$/$L^2$ data fidelity
• total variation minimization

2010 AMS SUBJECT CLASSIFICATIONS:

• 94A08
• 68U10
• 65K10