Unrolled Null Space Tuning dictionary learning for denoising

摘要

We investigate a dictionary learning approach in the context of image denoising via sparse coding, where the dictionary is adapted for Null Space Tuning (NST) recovery algorithms. We formulate a modified optimization problem for NST dictionaries and we propose a variant of the Iterative Shrinkage/Thresholding (ISTA) ISTA and Learned-ISTA iterations for learning it. The resulting model is evaluated in the context of image denoising with Deep K-SVD. Simulation results show faster convergence and improved efficiency, at least in the context of smaller training datasets.