Cosine-driven non-linear denoising

The Perona-Malik model is an effective but ill-posed model for denoising digital images by anisotropic diffusion. Instead of complex regularizations, we propose a new continuous model which is well-posed and show that it is nevertheless effective for denoising. In addition, an extension of our model offers the possibility of inducing a convergence for the discretization. A comparison to the original Perona-Malik model is carried out using an human vision-centered quality index which shows the improvements of our model when it comes to denoising.