Joint group sparse coding and weighted nuclear norm for Poisson denoising

It is difficult to remove Poisson noise because of its multiplicative and signal-dependent nature. In this paper, a new Poisson denoising model based on nonlocal self-similarity is introduced. It combines the weighted nuclear norm and group sparse coding as a regularization term, and makes full use of the low rank and sparse properties of similar image patches. Numerically, incorporating singular value decomposition and the variable splitting method, an alternating minimization method with an adaptive parameter selection strategy is proposed to resolve the new denoising model. Extensive experiments indicate that the proposed model outperforms the existing state-of-the-art Poisson denoising methods.