Hybrid patch similarity for image denoising

Presented is a new patch-based denoising method based on hybrid similarity, combining structure similarity and homogeneity similarity. The proposed method makes a good balance between the two similarity measures, using homogeneity similarity to enhance the patch regularity assumption, while keeping the block matching accuracy by structure similarity. Moreover, the use of hybrid similarity enables the use of large patch size to improve the robustness of the similarity measure. The results demonstrate the superiority of hybrid similarity over single similarity.