具有显著异常值的夜间模糊图像非盲去模糊  

Non-blind deblurring of night blurred images with significant outliers

作　　者：李振翮 武友新[1] LI Zhenhe;WU Youxin(School of Information Engineering,Nanchang University,Nanchang 330031,China)

机构地区：[1]南昌大学信息工程学院,江西南昌330031

出　　处：《南昌大学学报（工科版）》2020年第3期300-306,共7页Journal of Nanchang University(Engineering & Technology)

基　　金：江西省科技计划基金资助项目(20151BBE50065)。

摘　　要：为了抑制夜间模糊图像中常出现的各类异常值对去模糊的影响,提出了一种基于联合稀疏与最大熵先验的异常值检测与去模糊算法,该算法能够有效检测分布具有稀疏性的异常值。另一类异常值-饱和像素的分布通常具有聚集性,为此设计了一种高度饱和区域周边振铃伪影修正算法作为复原图像的后处理步骤。在包含冲击噪声的数据集上测试表明,本文算法能够对不同密度的噪声有效自适应。在部分饱和模糊图像上的实验显示,本文算法复原图像的平均SSIM值相比其他先进算法提高了0.1以上。In order to suppress the influence of various types of outliers that often appear in night blurred images,an outlier detection and deblurring algorithm based on joint sparse and maximum entropy prior was proposed,which could effectively detect outliers with sparse distribution.The distribution of another type of outlier-saturated pixels was usually clustered.Therefore,a correction algorithm for ringing artifacts around highly saturated areas was designed as a post-processing step to the restored image.Test results on the dataset with impulse noise showed that the algorithm can effectively adapt to different density of noise.Experiments on partially saturated blurred images showed that the average SSIM of our restored image was improved by about 0.1 higher than that of other advanced algorithms.

关 键 词：最大熵先验 稀疏先验 非盲去模糊 饱和像素 非高斯噪声 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]