A weighted total variation based image denoising model using mean curvature

A number of variational models for image denoising have been proposed in the last few years in order to improve the denoising performance of the classical Rudin-Osher-Fatemi model. In this work, we propose an adaptive model which uses the mean curvature of an image surface to control the strength of smoothing. A fast method for noise level estimation is proposed to improve the effectiveness of the proposed model. We analytically study the convergence of the model. In addition, we also provide a solver based on split Bregman method for numerical realization. Numerical experiments demonstrate that the proposed model yields good performance compared with high-order variational models. © 2020 Elsevier GmbH