Multiplicative Noise Removal via Nonlocal Similarity-Based Sparse Representation

Based on the sparse representation and by connecting the local and nonlocal regularizer, we proposed a new model to remove multiplicative noise in this paper. We first translated the multiplicative noise into additive noise by a logarithmic transformation, and then introduced a local regularizer based on dictionary learning and a nonlocal regularizer with nonlocal similarity to capture texture and edge information. A surrogate function-based iterative shrinkage algorithm was designed to solve the proposed model. Finally, the solution was transformed back into the real domain via an exponential function and bias correction. Experiments show that the denoised results of our model outperform state-of-the-art algorithms in terms of objective indices and subjective visual effect.