基于自适应超完备稀疏表示的图像去噪方法.doc
文本预览下载声明
基于自适应超完备稀疏表示的图像去噪方法*
肖 泉1,丁兴号1,王守觉1,2,郭东辉1,廖英豪1
(1 厦门大学信息科学与技术学院 厦门 361005;2 中国科学院半导体研究所 北京 100083)
摘 要:基于超完备字典的图像稀疏表示是一种新的图像表示理论,利用超完备字典的冗余性可以有效地捕捉图像的各种结构特征,从而实现图像的有效表示。当前稀疏表示的理论研究主要集中在稀疏分解算法和字典构造算法两方面。本文提出一种新的超完备字典构造算法:K-LMS算法,该算法由K均值聚类算法泛化获得,可用于超完备字典的自适应更新,以实现图像的有效表示。针对图像去噪问题,本文给出一种基于超完备稀疏表示的去噪方法,该方法利用图像在超完备字典上的自适应稀疏分解,通过阈值处理的方法实现了图像去噪,实验结果证实了本文所提方法的有效性。
关键词:稀疏表示;超完备字典;图像去噪;阈值
中图分类号:TN911.73 文献标识码:A 国家标准学科分类代码:510.4050
Image denoising based on adaptive over-complete sparse representation
Xiao Quan1, Ding Xinghao1, Wang Shoujue1,2, Guo Donghui1, Liao Yinghao1
(1 School of Information Science and TechnologyXiamen University, Xiamen 361005, China;2 Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China)
Abstract:The sparse representation based on over-complete dictionary is a new image representation theory. The redundancy of over-complete dictionary can make it effectively capture the geometrical characteristics of the images. Recent activities in this field concentrate mainly on the study of sparse decomposition algorithm and dictionary design algorithm. In this paper we propose a novel dictionary design algorithm, the K-LMS algorithm. It was obtained from generalizing the K-Means clustering algorithm and can be used in adaptive updating of over-complete dictionary in order to achieve sparse signal representations. Aiming at image denoising, a method based on over-complete sparse representation theory is introduced. With the application of image sparse representation in over-complete dictionary, the proposed algorithm reconstructs a simple threshold to realize image denoising. Experimental results demonstrate the effectivity of the proposed method.
Key words:sparse representation; over-complete dictionary; image denoising; threshold
1 引 言
图像信息的“有效”表示是图像去噪等处理任务的基础。“有效”表示是指能够用较少的系数捕获感兴趣目标重要信息的能力,即稀疏表示能力[1]。小波变换能够有效捕获点状奇异特征,对于一维含奇异点信号是一种最优的表示方法。二维图像信号中包含边缘等一维奇异特征,用小波表示图像时,要用点状奇异性来逼近线状奇异性,随着尺度的增加必然导致系数的增加,小波稀疏表示能力急速下降。为此,Donoho等学者提
显示全部