Image denoising via sparse representation using rotational dictionary

Abstract

In this paper, an image denoising algorithm is proposed using a rotational dictionary learned from image patches. Since the traditional dictionary-learning-based methods seldom take into account the rotational invariance for the dictionary, an improved K-means singular value decomposition (K-SVD) algorithm is developed with the rotation of atoms. In our method, the rotational version of atoms is introduced to greedily match the noisy image in sparse coding procedure. On the other hand, in dictionary learning procedure, to maximize the diversity of atoms, a rotational operation on residual error is adopted such that the rotational correlation among atoms is removed. As the novel strategy exploits the rotational invariance of atoms, more intrinsic features existing among image patches can be effectively extracted. Experiments also illustrate that the proposed method can achieve better performance than some other well-developed denoising methods.