小波变换在图像降噪中的应用研究

【作者】 矫媛；

【导师】 赵志刚； 郭振波；

【作者基本信息】 青岛大学 ， 计算机应用技术， 2008， 硕士

【摘要】 在图像处理过程中,图像的采集、转换和传输常常受到成像设备与外部环境噪声干扰等影响,产生降质。图像噪声对数字图像的后续处理影响较大,因此对图像噪声的去除有很重要的现实意义。小波分析是国际上新兴的一个前沿研究领域,随着小波理论的不断完善,小波分析在图像去噪中也得到了广泛的应用。本文首先介绍了近年来小波分析的发展及其在图像处理方面的应用情况,然后系统地描述了目前常用的小波图像去噪方法,并对这些算法进行了比较,分析了各算法的基理、特性以及存在的不足。本文在对小波理论和图像去噪算法研究的基础上,提出了三种新的小波图像降噪方法:基于平稳小波变换的邻域依赖自适应软阈值图像降噪方法、基于尺度连贯性边缘检测的邻域依赖小波收缩图像去噪方法和基于上下文模型的小波包图像降噪方法。其中第一种方法,充分考虑了尺度间与尺度内小波系数的依赖关系,并利用平稳小波变换的信息冗余性来去除噪声;第二种方法侧重于在去除噪声的同时保持边缘,将检测到的边缘点与非边缘点对应的小波系数区别对待,分别用不同的方法进行处理,采用不同的收缩因子进行收缩;而第三种方法则利用了小波包变换更加精细的分解,通过上下文模型估计每一个小波包系数的边缘方差,自适应调整阈值,得到了令人满意的去噪效果。最后,通过仿真实验,验证了本文提出的三种图像去噪方法的有效性。更多还原

【Abstract】 In the course of image processing,the collection,transformation and transmission of images are frequently affected by imaging equipments and noises in exterior environment, therefore,image quality declines.Because noises have big infection to the continuous processing of images,it has very important practical meaning to noises reduction. Wavelet analysis is a new international research field and along with continuous perfection of wavelet theory,it is also widely used in image denoising field.The development of wavelet analysis and its application situation in image processing are firstly introduced.Then,wavelet image denoising methods which are often used at present are described in detail.These algorithms are compared and the principle, characteristic and disadvantage of each algorithm are also analyzed.This paper chooses wavelet analysis-this new mathematic tool and on the basis of research on wavelet theory and image denoising algorithms,three new wavelet denoising methods are proposed.They are adaptive soft threshold image de-noising based on stationary wavelet transform and neighbor dependency,wavelet shrinkage image de-noising method based on scales consistency edge detection and neighbor dependency and wavelet packet image denoising method based on context model.The first method fully considered dependency relation among scales and within scales wavelet coefficients. The information redundancy of stationary wavelet transform is also used to reduce noises. The second method emphasized edge preservation while eliminating noises.The edge pixels and non-edge pixels which are detected are treated differently and disposed in different methods.That is to say different shrinkage factors are used on them.While the third method made use of the more elaborated decomposition of wavelet packet and through context model,each wavelet packet coefficient’s edge variance is estimated and then,threshold is adaptively adjusted.The denoising effects are satisfied.At last,through simulation experiments,the availabilities of these three new denoising methods are validated.更多还原