Iterative decoupling deconvolution network for image restoration

Yixing Ji, Shengjiang Kong, Weiwei Wang, Xixi Jia, Xiangchu Feng

Abstract: Highlights•A deep unfolding model is proposed by unfolding the iterations of IDDBM3D.•Two CNN modules are designed to replace the analytical analysis/synthesis transform.•The CNN modules can learn features more effectively than the analytical ones.•A new denoiser module is designed to replace the soft thresholding operator.•The denoiser removes the unknown noise in the intermediate images effectively.