摘要
针对大气湍流引起的红外图像模糊问题,提出一种基于混合正则化的模糊核估计模型。根据图像主要边缘的稀疏性,采用图像梯度的L_0范数为正则化项;通过分析模糊核的特性,提出能适用于复杂模糊情况的核L_0-L_2范数正则化约束。复原模型的优化过程中,结合变量分裂策略和增广拉格朗日法交替估计图像和模糊核,并利用快速傅里叶变换,实现模糊核的快速、准确估计;最终根据估计的模糊核,复原得清晰图像。实验结果表明,本文算法可以更好地复原退化图像,在主观视觉和客观质量评价方面都有所提高。
A blur kernel estimation method based on hybrid regularization is proposed to restore the blurred infrared image caused by turbulence. Firstly,according to the sparsity of the prominent edges and the smoothing of homogenous regions in the natural image, a L0-norm of image gradient is as regularization item. Through analyzing the feature of the blur kernel,a L0 - L2-norm regularization constraint is applied to the complex blur situation. Secondly, the image and the blur kernel were estimated ahernately by the variate splitting and augmented Lagrangian method in the optimization procedure of restoration model. The fast and accurate estimation of the blur kernel was achieved by FFT, Finally,the clear image was restored by the estimated blur kernel. Experimental results demonstrate that the proposed algorithm can better restore the degraded image,and the subjective vision and the objective measuremeut have been improved.
出处
《激光与红外》
CAS
CSCD
北大核心
2017年第7期884-888,共5页
Laser & Infrared
基金
国家自然科学基金项目(No.61175120)资助
关键词
盲复原
L0正则化
增广拉格朗日法
湍流退化
红外图像
blind restoration
L0-norm regularization
augmented Lagrangian
turbulence-degraded
infrared image
