Image Restoration using Denoising Autoencoder Priors

Siavash Arjomand Bigdeli1 Matthias Zwicker1
1Computer Graphics Group, University of Bern

Technical report

Teaser
We propose a natural image prior based on a denoising autoencoder, and apply it to image restoration problems like nonblind deblurring. The output of an optimal denoising autoencoder is a local mean of the true natural image density, and the autoencoder error is a mean shift vector. We use the magnitude of the mean shift vector as the negative log likelihood of our prior. To restore an image from a known degradation (left column, known blur and noise), we use gradient descent to iteratively minimize the mean shift magnitude while respecting a data term. Hence, step-by-step (second column from left to right) we shift our solution closer to its local mean in the natural image distribution.

Abstract

We propose to leverage denoising autoencoder networks as priors to address image restoration problems. We build on the key observation that the output of an optimal denoising autoencoder is a local mean of the true data density, and the autoencoder error (the difference between the output and input of the trained autoencoder) is a mean shift vector. We use the magnitude of this mean shift vector, that is, the distance to the local mean, as the negative log likelihood of our natural image prior. For image restoration, we maximize the likelihood using gradient descent by backpropagating the autoencoder error. A key advantage of our approach is that we do not need to train separate networks for different image restoration tasks, such as non-blind deconvolution with different kernels, or super-resolution at different magnification factors. We demonstrate state of the art results for non-blind deconvolution and super-resolution using the same autoencoding prior.

Additional Information