Accelerated and enhanced multiplicative deblurring schemes

Abstract

We propose modifications of the Richardson–Lucy iterations (RL) and the Image Space Reconstruction Algorithm (ISRA) that demonstrate accelerated convergence and lead to improved image restoration results. We show that the iterations of RL, ISRA, and the proposed modifications can be interpreted as fixed-point iterations corresponding to the minimizers of certain variational problems. We demonstrate that combining each iteration of the proposed modifications with an adaptive image smoothing procedure leads to substantial improvements of the image restoration results. An implementation is available at https://github.com/fayolle/Mult_BBDeblur_demo.