Hybrid regularization based restoration algorithm for sparse images

Su Xiao,Guoqiang Han
DOI: https://doi.org/10.12733/jcis8491
2013-01-01
Journal of Computational Information Systems
Abstract:Image restoration, as a linear inverse problem, should be regularized to obtain desired results. However, single regularization may not enable restored images to demonstrate multiple characteristics. Therefore, to facilitate image restoration and to impose desirable characteristics on the results, a novel imagerestoration algorithm for spare images is proposed. To address the image-restoration problem, a new model adopting hybrid regularization is initially established. To solve this model efficiently and easily, several constraints based on variable splitting are introduced. The alternating optimization procedure is adopted to decompose the constrained image-restoration model into several equivalent unconstrained subproblems. Moreover, by alternately solving these subproblems based on the proximal mapping method, the original image-restoration problem can be efficiently solved. In the simulated experiments, the proposed algorithm successfully restores three kinds of observed images, thus proving its effectiveness. Compared with the proposed algorithm with single regularization, the proposed algorithm with hybrid regularization obtains better results. Copyright © 2013 Binary Information Press.
What problem does this paper attempt to address?