摘要
运动模糊图像盲复原是图像处理中的关键问题之一.由于模糊信息估计的复杂性以及图像噪声的影响,现有算法往往难以做到高质量的图像复原.为改善模糊信息估计的效果,提出一种基于自适应线性滤波的改进算法.首先在原有模糊信息估计过程中引入自适应动态线性滤波以抑制噪声影响,达到改善模糊信息估计结果的目的,同时可以起到调整优化目标的作用,使原问题变得较容易求解,从而获得高质量的模糊信息估计;在此基础上提出了改进的重定权值split Bregman迭代法,用于获得模糊信息后求解原始图像的过程中,进一步改善模糊图像复原的效果.实验结果表明,与3种现有的模糊图像盲复原算法相比,该算法能更准确地估计模糊信息,对多数图像复原任务具有更好的鲁棒性,能有效地用于运动模糊图像复原任务.
We propose a novel approach to estimating blur kernel in the Blind Image Deblurring.This is a challenging problem,because image restoration with unknown kernels is an ill-posed deconvolution process.Existing methods are also sensitive to image noise and compression artifacts.Our method overcomes these drawbacks by introducing an adaptive linear filter to handle image noise.The blur kernel is automatically learned together with the adaptive linear filter simultaneously.Then,the clear images are restored using an improved non-blind deblurring method based on reweighted split Bregman iteration optimization.Compared to the state of the arts,our approach is more robust to image noise and shows stable performance in single image deblurring tasks.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2014年第3期457-464,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61233011,90820306)
