【作者】 龚晓彦；

【导师】 郭希娟；

【作者基本信息】 燕山大学 ， 计算机应用技术， 2010， 硕士

【摘要】 医学图像配准是现代医学图像处理的关键技术,为医生提供了科学的诊断依据。基于互信息的配准方法具有配准精度高、鲁棒性强等优点,近年来已成为医学图像配准领域的研究热点。本文研究了大量基于互信息医学图像配准的算法,所做的主要工作有以下几个方面:首先,介绍了医学图像配准的研究背景及发展现状,深入研究了基于互信息配准的相关理论知识,分析了算法的原理及配准流程,并对基于互信息的配准方法存在的优缺点及改进方法进行了详细的讨论。其次,在互信息理论基础上,对图像插值技术进行了研究,针对传统插值方法插值精度低和插值速度慢的缺点,提出了一种新的插值算法,即基于像素点的亮度绝对误差的图像插值算法。再次,针对基于互信息配准方法存在局部极值、配准时间长等问题,提出利用小波变换中的多分辨率思想,先对待配准图像进行多分辨率分解,在低分辨率下采用FGA优化算法取得配准的初步结果,再在高分辨率下采用Powell优化算法取得精确结果。最后,采用美国Vanderbilt大学RREP项目组提供的国际通用图像数据,利用Matlab7.0工具对基于互信息的医学图像配准的两种新算法进行单模和多模配准仿真实验,实现了互信息法的图像配准,同时还将配准结果进行比较分析,从而验证了新算法的精确性和鲁棒性。更多还原

【Abstract】 Medical image registration is the key technology of the modern medicine; it provides the scientific diagnosis for doctors. Registration based on Mutual Information are popular in medical image registration processing fields, which are accepted as the most accurate and robustness. In this paper, a large number of Registration based on Mutual Information algorithms are researched. The main results in this paper are as follows:Firstly, the background knowledge and current situation were introduced, and the correlative theoretical knowledges Registration based on Mutual Information were deeply researched, and the principle of mutual information and registration processes were introduced, and the strengths and weaknesses of Registration based on Mutual Information algorithms and improving methods were discussed in detail.Secondly, on the basis of mutual information, image interpolation technology are researched, aim at the traditional interpolation method has low accurate and slow speed, a new interpolation algorithm are proposed, that is, the image interpolation algorithm based on the absolute error of the brightness of pixels, which combines the advantages of nearest interpolation and bicubic interpolation, without prejudice to the case of interpolation accuracy can greatly improve the speed of image interpolation.Thirdly, to overcome Registration based on Mutual Information algorithms has the shortcoming of the local maxima and the computing expansive, wavelet- based multi-resolution approach was applied in registration. The sub-images which were decomposed by wavelet transformation were registered using FGA, and then the results were applied as the start point for the registration of high-resolution image using Powell’s method. Finally, the single-mode and multimode registration experiments of two algorithms were done under Matlab7.0 environment, in which use the image data which provided by international standard on the United States Vanderbilt University RREP project team, which realized image registration based on mutual information, at the same time, the results of registration are comparative analysis, and the accuracy and robustness of the new algorithm were verified.更多还原