关键词: 图像盲卷积, 稀疏表示, 图像梯度, 超拉普拉斯先验, 图像去模糊

Abstract: In view of the limited ability in detail preserving of sparse representation-based blind image convolution algorithm， this paper proposes a blind image convolution algorithm based on sparsity constraints of image patches and gradient prior. Although each image patch can be approximated by a sparse linear combination of atom signals in an over-complete dictionary, it brings about the artificial effects among image blocks. In order to solve this problem, the image gradient prior and Hyper-Laplacian priors are incorporated into the sparse representation blind convolution model to estimate the blur kernel. Once the blur kernel is known, we can apply the non-blind deconvolution algorithm to obtain the latent image, which brings about some ringing effects. Therefore we utilize the Hyper-Laplacian priors to recover the final latent image. Experimental results demonstrate that the proposed improved method can remove artifacts and render better deblurring images, compared to other state-of-the-art deblurring methods.

Key words: blind image convolution； sparse representation； image gradient； Hyper-Laplacian priors, image deblurring 