DeepRED: Deep Image Prior Powered by RED

25 Mar 2019  ·  Gary Mataev, Michael Elad, Peyman Milanfar ·

Inverse problems in imaging are extensively studied, with a variety of strategies, tools, and theory that have been accumulated over the years. Recently, this field has been immensely influenced by the emergence of deep-learning techniques. One such contribution, which is the focus of this paper, is the Deep Image Prior (DIP) work by Ulyanov, Vedaldi, and Lempitsky (2018). DIP offers a new approach towards the regularization of inverse problems, obtained by forcing the recovered image to be synthesized from a given deep architecture. While DIP has been shown to be quite an effective unsupervised approach, its results still fall short when compared to state-of-the-art alternatives. In this work, we aim to boost DIP by adding an explicit prior, which enriches the overall regularization effect in order to lead to better-recovered images. More specifically, we propose to bring-in the concept of Regularization by Denoising (RED), which leverages existing denoisers for regularizing inverse problems. Our work shows how the two (DIP and RED) can be merged into a highly effective unsupervised recovery process while avoiding the need to differentiate the chosen denoiser, and leading to very effective results, demonstrated for several tested problems.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution Set14 - 4x upscaling DeepRED PSNR 27.63 # 69
Image Super-Resolution Set14 - 8x upscaling DeepRED PSNR 24.28 # 7
Image Super-Resolution Set5 - 8x upscaling DeepRED PSNR 26.04 # 7

Methods


No methods listed for this paper. Add relevant methods here