Non-local sparse models for SAR image despeckling

This paper propose a non-local sparse model for SAR image despeckling. Sparse coding models and non-local means have been both proven very effective in natural image restoration tasks. While self-similarities exist widely in SAR images, which encourages combining these two approaches together for SAR image despeckling tasks. A grouped-sparsity regularizer is imposed to enforce similar image patches to admit similar estimates. Image adaptive dictionary is learned by block-coordinate descent algorithm. Considering the importance of point targets, a new term is integrated into sparse coding models for preserving of point targets. Experimental results show the effectiveness of the proposed algorithm in SAR image despeckling task.