摘要
通过对现有的二维盲图像恢复算法的探讨,提出了两种基于L1双正则化的二维盲图像恢复算法。一种是最小化L2-L1代价函数,为了实现边缘保持和噪声抑制;另一种是通过最小化L1-L1代价函数来处理非高斯噪声的情况。所提的算法是一种广义的梯度算法,它通过引入绝对值函数的弱导数来处理不可微的情况。实验结果表明,与NAS-RIF算法和DR算法相比,所提出的两种二维算法能够更快速地获得好的图像估计。
By exploring the existing two-dimensional （2-D） algorithms for blind image restoration, this paper proposes two new two-dimensional algorithms for blind image restoration based on an L1 double regularization approach. One is formulated as the minimization of a L2-L1 cost function to achieve edge preservation and noise suppression. The other is viewed as the minimization of a L1-L1 cost function for blind image restoration under non-guassian noise environments. Thus a generalized gradient algorithm is introduced by using a weak derivative of the absolute value function to deal with the non-differentiable case. Experimental results show that the proposed two-dimensional algorithms can obtain a better restored image and the estimated PSF with a faster speed than both the NAS-RIF algorithm and the DR algorithm.
出处
《三明学院学报》
2014年第2期6-13,共8页
Journal of Sanming University
基金
龙岩学院服务海西面上项目(LYXY2011071)
关键词
盲图像恢复
L1双正则化方法
二维实现算法
广义梯度算法
blind image restoration
L1 double regularization approach
2-D implementation algorithm
generalized gra- dient algorithm
