Deep Mean-shift prior

Deep Mean-shift prior is an approach for image restoration developed based on the estimate of the natural image probability distribution smoothed by a Gaussian kernel.

It is created as a Bayes estimator that boost the Gaussian smoothed density of natural images. Its framework is an example of Bayesian inference using deep learning.

Deep Mean-shift prior allows users to solve noise-blind restoration problems. It builds on denoising

autoencoders (DAEs). It provides image restoration techniques based on gradient-descent risk minimization with competitive results for noise-blind image deblurring, super-resolution and demosaicing