Deep learning based low-rank prior ADMM for image recovered with Cauchy noise

Cauchy noise is a non-Gaussian-like noise with impulsive property, which results in difficulties in denoising solver design. By combining a flexible low-rank prior based on the nonlocal self-similarity of the natural image and an efficient deep image prior characterizing a generic learned from many images, we propose a novel model based on a plug-and-play framework containing mixed regular terms for removing additive Cauchy noise. We adopt deep learning based alternating direction method of multiplier (ADMM) algorithm to solve the proposed model. Numerical experiments illustrate that our method has better performance than other state-of-the-art methods in terms of visual quality and quantitative metrics PSNR/SSIM.