Image De-noising Based on Improved Sparse Gradient Constraint

Sparse representation model adopts an image patch as a linear combination of a few atoms chosen out from an overcomplete dictionary, and they have shown promising results in various image restoration applications. Gradient feature is important structure information in image processing. Their combination is not sensitive to noise, which can improve and enhance the accuracy of similarity measure, especially in strong power noise. Non-local means (NLM) algorithm can obtain very good denoising results by making full use of the self-similarity and redundancy information. However, the weight function of NLM algorithm cannot accurately measure the similarity between image patches in the case of strong noise. Therefore, sparse gradient is introduced into NLM algorithm so as to obtain better performance in both objective evaluation and visual effect. Firstly, differing from the traditional local gradient algorithm, a global sparse gradient model is introduced to propose an adaptive sparse gradient algorithm. Thus, it can be solved by forward-backward divided algorithm. In addition, the weight function of NLM algorithm can be improved by adopting sparse gradient mode. Experimental results showed that, compared with some algorithms based on gradient information, our proposed algorithm not only enhances the objective evaluation, but also improves the visual effect.