盲图像恢复的两个关键问题研究.pdfVIP

中国科技论文在线 盲图像恢复的两个关键问题研究# 黄丽清，夏又生** （福州大学数学与计算机科学学院，福州 350100 ） 5 摘要：正确估计模糊核尺寸和识别模糊核类型是盲图像恢复的关键问题。为了解决这两个关 键问题，首次提出了一种基于特征线检测的模糊核尺寸估计算法；提出了一种基于字典学习 的模糊核类型识别算法。 特征线检测主要包括图像对数归一化特征，二进值转化及特征线 统计规则。字典学习包括输入观测图像的结构相似性指标，训练库图像的相似性指标，判定 10 结构最相似的类型图。实验结果证实所提出的两种算法能很有效地估计模糊核尺寸和识别模 糊核类型。最后，应用结果进一步显示所提出的两种技术对提高盲图像恢复效果能起到重要 作用。 关键词：盲图像恢复；模糊核；尺寸估计;类型识别 中图分类号：TP319 15 Two Important Problems of Blind Image Restoration Huang Liqing, Xia Yousheng (College of Mathematics and Computer Science,FuZhou University ,FuZhou 350100) Abstract: Effectively estimating the size of blur kernel and identifying its type are two key 20 issues of blind image restoration. In order to solve them, a feature line detection-based algorithm for estimating the size of blur kernel and a dictionary learning-based algorithm for identifying the type of blur kernel are developed in this paper, respectively. The feature line detection mainly includes logarithmic normalized feature matrix of blurred images, binary transform of feature matrix, and feature line estimation. Dictionary learning includes logarithmic normalized 25 feature matrix of blurred images from observed image and dictionary images, the calculation of structural similarity index, finding the best structural similarity between observed image and dictionary images. Experimental results show that the proposed two algorithms are very effective. Furthermore, an application to blind image restoration shows that the proposed two techniques are useful and necessary. 30 Key words: blind image restoration;blur kernel; size estimation, type recognition 0 引言