Low Rank Poisson Denoising (LRPD): A Low Rank Approach Using Split Bregman Algorithm for Poisson Noise Removal From Images

Prashanth Kumar G., Rajiv Ranjan Sahay; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Occurrence of Poisson noise in captured observations is inevitable in various real imaging applications ranging from medical imaging to night vision imaging. Restoration of fine details of an image is difficult when it is corrupted by Poisson noise. Recently, low rank approaches outperformed several state-of-the-art techniques for image denoising, deblurring, image completion, super-resolution, etc. The ability of low rank techniques to preserve fine details, even though the image is corrupted by severe noise, motivated us to develop an optimization framework wherein, we propose to use a low rank prior for Poisson noise removal. In the proposed low rank Poisson denoising (LRPD) algorithm, we resort to split Bregman technique to solve an appropriate objective function. We incorporate the forward-backward splitting scheme to minimize the first subproblem and the weighted nuclear norm minimization (WNNM) for the second subproblem of split Bregman algorithm to arrive at the final solution. We conduct several experiments on both simulated and real-world Poisson noisy data and show the superiority of the proposed method over other state-of-the-art Poisson denoising techniques.