Sparse Coding based Robust Image Denoising via Coupled Dictionary

This paper addresses the image denoising problem using the recent advancements in sparse representation based image restoration. This novel approach is based on sparse coding using a pre-learned coupled dictionary. Prior research suggests that it is possible to represent every image patch as a sparse linear combination of dictionary elements learned from a set of example patches. Joint training of coupled dictionaries for clear and noisy images helps to enforce self-similarity between noisy and latent images. The sparsity measure over such a coupled dictionary is used as a regularisation constraint to recover a denoised image patch from a noisy image patch. The experimental results indicate that the proposed method appears to be competitive with other denoising methods.