A General Iterative Regularization Framework For Image Denoising

Michael R. Charest, Michael Elad, Peyman Milanfar

Abstract: Many existing techniques for image denoising can be expressed in terms of minimizing a particular cost function. We address the problem of denoising images in a novel way by iteratively refining the cost function. This allows us some control over the tradeoff between the bias and variance of the image estimate. The result is an improvement in the mean-squared error as well as the visual quality of the estimate. We consider four different methods of updating the cost function and compare and contrast them. The framework presented here is extendable to a very large class of image denoising and reconstruction methods. The framework is also easily extendable to deblurring and inversion as we briefly demonstrate. The effectiveness of the proposed methods is illustrated on a variety of examples.