Learning Nonlinear Spectral Filters for Color Image Reconstruction

Michael Moeller, Julia Diebold, Guy Gilboa, Daniel Cremers, “Learning Nonlinear Spectral Filters for Color Image Reconstruction”, ICCV 2015.


This paper presents the idea of learning optimal filters for color image reconstruction based on a novel concept of nonlinear spectral image decompositions recently proposed by Gilboa. The general idea is to use total variation regularization along with Bregman iterations to represent the input data as the sum over image layers containing features at different scales. Filtered images can be obtained by weighted linear combinations of the different frequency layers. We show that learning the optimal weights can significantly improve the results in comparison to the standard variational approach, and can achieve state-of-the-art results. While we focus on the problem of image denoising, our general framework extends to a number of image reconstruction tasks.