Non-local total variation regularization models for image restoration

Restoration of images corrupted by data-correlated Rayleigh noise distribution has not been studied much extensively in the literature, unlike the other noise distributions. In this paper, we analyze the degradations due to a data-correlated Rayleigh noise and a linear blurring artifact. This work employs a variance stabilization approach and two variational approaches for restoring images from their noisy and blurred observations. The split-Bregman iterative scheme is used for numerically solving the models to improve their convergence rates. Furthermore, non-local total variation and non-local total bounded variation priors are being used as regularizers in these models to improve their restoration efficiency. Various synthetic and real images (such as ultrasound and synthetic aperture radar images) are tested to show the performance of these models.