Hyper-Laplacian Regularized Non-Local Low-Rank Prior for Blind Image Deblurring

• Published in: IEEE access Vol. 8; pp. 136917 - 136929
• Main Authors: Chen, Xiaole, Yang, Ruifeng, Guo, Chenxia, Ge, Shuangchao, Wu, Zhihong, Liu, Xibin
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Approximation algorithms; Blind deblurring; Estimation; hyper-laplacian; Ill posed problems; Image edge detection; Image restoration; Iterative methods; Kernel; low-rank matrix approximation; Matrix decomposition; non-local self-similarity; non-uniform deblurring; Optimization; Self-similarity
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2020.3010540

Blind deblurring of single image is a challenging image restoration problem. Recent various image priors have been successfully explored to solve this ill-posed problem. In this paper, based on the non-local self-similarity, we propose a novel method for blind image deblurring, which can simultaneously capture the intrinsic structure correlation and spatial sparsity of an image. Specifically, we use the hyper-Laplace prior to model the structure information of non-local similar patches, and embed it into the low-rank model as a smooth term of the energy equation. Since the established energy function is non-convex, an effective iterative optimization scheme is designed to effectively implement the proposed algorithm. In addition, we evaluate the proposed method for non-uniform deblurring problem. Extensive experimental results on both synthetic and real-world images show that the proposed method performs competitively against the state-of-the-art methods.