Abstract
The total-variation regularization is an effective tool in image processing. Nevertheless, applying the TV regularization to an image contaminated by a large quantity of noise, it will produce a denoised image in which speckled colored-noise is noticeable. This color artifact is caused by the discretization scheme of the TV regularization term, in which simple difference operations between neighboring two pixels are used. We introduce a new multi-resolution scheme into the discretization of the TV regularization term, in which the TV norm is evaluated and minimized at the each resolution image. We present the simulation results and demonstrate superiority of our new approach.
