Unsupervised learning of denoisers with compressive sensing measurements

Abstract

Recently, deep learning based compressive recovery methods have been proposed and have yielded state-of-the-art performances. Ironically, training deep neural networks for them requires “clean” ground truth, but obtaining the best quality images from undersampled data requires well-trained deep networks. To resolve this dilemma, we propose methods that are able to train deep denoisers from undersampled measurements without ground truth. Our methods yielded comparable performances to the methods with ground truth for various image recovery problems.