A regularization parameter selection model for total variation based image noise removal

摘要

Total variation regularized model is a powerful tool in image noise removal due to its edge-preserving property of an image. One important procedure in the model is to determine the regularization parameter which has an important role in balancing the data-fidelity and the regularity of the denoised image. Discrepancy principle is a classical method for selecting the regularization parameter, which provides an upper bound for the value of the data-fitting term. For the regularization parameter, it is easier to estimate its upper bound as the statistical property of the noise is known. The contribution of this paper is twofold. First, we propose an iterative algorithm to estimate an optimal upper bound by applying the consistency between the value of data-fitting term and the upper bound. Second, we develop a dual-based method to solve the constrained problem which can avoid the computation of the Lagrangian multiplier associated with the constraint. The new algorithm can simultaneously solve the solution of the constrained problem and the estimate of the regularization parameter. Numerical results are given to show that the proposed algorithm is better than some state-of-the-art methods in both efficiency and accuracy. (C) 2018 Elsevier Inc. All rights reserved.