Multiplicative noise removal using deep CNN denoiser prior

Abstract

Multiplicative noise removal is always a hard problem in fundamental image processing task. Many methods are proposed for the multiplicative noise removal by using different denoiser prior in variational framework. Among the image prior, total variation (TV) are first proposed and then many other regularization such as PM, TGV, nonlocal and many other priors are also proposed for enhance the denoising ability. Although using the priors can get good performance, the models are hard to be resolved with sophisticated priors. A new model based on the deep CNN denoiser prior for removing multiplicative noise is proposed in this paper. The proposed energy function is effectively calculated via several sub-optimal questions by split bregman method and alternative minimization is used for the solution. The proposed method needn't deduce the sophisticated formula and can achieve good performance. From the experiments, we can see that our method achieved good results.