【作者】 雷浩鹏；

【导师】 李峰；

【作者基本信息】 长沙理工大学 ， 计算机应用技术， 2010， 硕士

【摘要】 数字图像的噪声主要来源于图像采集和传输的过程,是干扰图像质量最主要的因素之一,同时还影响着人们对图像信息的读取以及对图像所做的后续处理,所以图像去噪问题一直是数字图像研究领域中的一个研究热点。本文对图像去噪的一些关键技术和算法进行了较为深入的探索和研究,所做的主要工作如下:首先介绍了数字图像去噪算法的发展概况和研究现状,在总结一些传统去噪算法的同时指出了去噪算法中几个最新的研究领域;然后提出了几种常见图像噪声的数学统计模型和去噪质量评判标准,接着介绍了两大传统图像去噪算法:中值滤波法和维纳滤波法,对它们的去噪原理和性能进行了详细的讨论与分析。其次研究了两种最新的图像分析工具:小波变换和多尺度几何变换理论,重点探讨如何将小波变换和Contourlet变换应用于图像去噪,分析了图像经过这两种变换后系数的分布特点以及阈值函数的选取原理,并在此基础上提出了一种基于Contourlet变换的分层阈值去噪算法;并且对以上算法进行了详细的仿真实验,分析了它们的去噪性能。最后针对原Contourlet变换中的缺陷,提出了一种改进的非采样Contourlet变换:多小波—非采样Contourlet变换:首先利用多小波对图像进行多尺度分解,再结合非下采样方向滤波器组进行方向分解。并将此变换应用于图像去噪之中,并根据分解所得到的各方向子带的关系,改进了Bayes Shrink自适应阈值取值方法,由此提出了一种基于多小波—非采样Contourlet变换的自适应阈值去噪算法。本文对该算法进行仿真实验,且与一些已有的去噪算法进行了对比,实验结果表明:应用此算法的图像去噪效果与传统算法相比,有了一定的提高,能够保留更多的图像细节信息,且具有一定的实用价值。更多还原

【Abstract】 Digital image noise mainly originates from the process of image acquisition and transmission, which also is one of the main factors influencing the quality of image information. Therefore image de-noising is a popular research subject in image processing area. This paper has researched some crucial techniques and algorisms deeply on this area. The main work and research results will be given as follows:Firstly this paper provides an overview of development and current situation on image de-noising, and proposes some latest research area, gives some common image statistical mathematical model and the standard of judgment of de-noising quality, then introduces two classical de-noising algorithms :media filter method and wiener filter method, analyses and discusses their de-noising principles and properties emphatically.Secondly this paper discusses two latest image analysis tools: wavelet transform and Multi-scale Geometry transform, discusses how to apply this two transform on image de-noising, analyses their spreading property of coefficients and threshold choosing principles; on this basic this paper proposes a hierarchical threshold de-noising method based on Contourlet transform, carries on simulation experiment to these algorithm and compares their de-noising results.Finally according to the drawbacks of Contourlet transform, this paper proposes an improved Contourlet transform: multi wavelet-nonsubsampled Contourlet transform, this algorithm uses multi wavelet for multi-scale decomposing and combines nonsubsampled filter banks for multi-direction decomposing, applies this transform on image de-noising and improves Bayes Shrink Adaptive Threshold method according to the relation between the decomposing sub bands. So this paper proposes a new algorithm which is called multi wavelet-nonsubsampled Contourlet transform de-noising algorithm. The experiment result shows this algorithm is superior to some traditional algorithm that can reserve more image detail information and get a better de-noising effect, which also has certain of application value.更多还原