关键词: 机器视觉/
贝叶斯框架/
联合双边滤波器/
图像盲复原
English Abstract
Fast Bayesian blind restoration for single defocus image with iterative joint bilateral filters
Yin Shi-Bai1,Wang Wei-Xing2,
Wang Yi-Bin3,
Li Da-Peng1,
Deng Zhen4
1.Department of Economic Information Engineering, Southwestern University of Finance and Economics, Chengdu 611130, China;
2.Department of Information Engineering, Chang'an University, Xi'an 710064, China;
3.Department of Engineering, Sichuan Normal University, Chengdu 610101, China;
4.Department of Information Engineering, Ningxia University, Yinchuan 750021, China
Fund Project:Project supported by the Major Program of the National Natural Science Foundation of China (Grant No. 91218301), the Young Scientists Fund of the National Natural Science Foundation of China (Grant No. 61502396), the Fundamental Research Fund for the Central Universities, China (Grant Nos. JBK150503, JBK160135), and the Natural Science Foundation of Ningxia, China (Grant No. NZ15054).Received Date:09 July 2016
Accepted Date:11 September 2016
Published Online:05 December 2016
Abstract:It is significant to realize effective defocus image restoration for acquiring clear image in military and geological examination field. Most of existing algorithms have the problems of large computational cost, ringing and noise sensitivity, hence a novel approach by iterative joint bilateral filtering under Bayesian framework is proposed. Firstly, it utilizes defocus image depth estimation to compute the point spread function in the Bayesian framework. Then a minimum optimization problem is built to represent the blind restoration problem. After inferencing the solution procedure of the minimum optimization problem, we find that the joint bilateral filters can be used to search the optimal solution, which not only simplifies the searching procedure but also reduces the computational cost. Finally, an iterative joint bilateral filtering is designed to realize the image restoration. That means that the original restored image obtained from the bilateral filtering is used to design the guide image for the joint bilateral filters, and the guide image will serve as the input of the optimization problem for acquiring the better optimal result. This procedure is repeated until convergence. The experimental results indicate that this method can yield the ringing, reduce the computational cost, and remove the noise. Generally speaking, the average pixel error of 85% images is under 0.03, which has improved 19% comparing with the same error rang of existing algorithms, and 78% shorter than those of compared algorithms. It can be used in the engineering practice of blind restoration for single defocus image.
Keywords: machine vision/
Bayesian framework/
joint bilateral filtering/
image blind restoration
