Zhang Kai Zuo Wangmeng Gu Shuhang Zhang Lei

Abstract

Model-based optimization methods and discriminative learning methods havebeen the two dominant strategies for solving various inverse problems inlow-level vision. Typically, those two kinds of methods have their respectivemerits and drawbacks, e.g., model-based optimization methods are flexible forhandling different inverse problems but are usually time-consuming withsophisticated priors for the purpose of good performance; in the meanwhile,discriminative learning methods have fast testing speed but their applicationrange is greatly restricted by the specialized task. Recent works have revealedthat, with the aid of variable splitting techniques, denoiser prior can beplugged in as a modular part of model-based optimization methods to solve otherinverse problems (e.g., deblurring). Such an integration induces considerableadvantage when the denoiser is obtained via discriminative learning. However,the study of integration with fast discriminative denoiser prior is stilllacking. To this end, this paper aims to train a set of fast and effective CNN(convolutional neural network) denoisers and integrate them into model-basedoptimization method to solve other inverse problems. Experimental resultsdemonstrate that the learned set of denoisers not only achieve promisingGaussian denoising results but also can be used as prior to deliver goodperformance for various low-level vision applications.