【作者】 薛辉；

【导师】 李雄飞；

【作者基本信息】 吉林大学 ， 计算机软件与理论， 2011， 硕士

【摘要】 近年来,医学图像融合日益成为图像处理、计算机视觉领域的研究重点,是计算机应用技术研究的重要领域。由于不同的医学图像采集设备具有不同的成像特点,使得采集到的不同的医学图像显示出不同的医学信息：CT图像侧重呈现了人体的骨质信息,而MRI图像则重点显示了人体的软组织信息。随着现代医学对诊断水平要求的不断提升,将CT和MRI等不同医学图像融合正呈现出前所未有的重要性,这将有利于对病人病灶部位的精度定位,以及排除采集设备自身的影像干扰,将有助于推动医疗诊疗水平的大幅提升。由美国工程院院士、美籍华人黄等人于1998年提出的一种全新的信号分析技术：Hilbert-Huang变换,它克服了传统的小波变换针对线性稳定信号的局限,同时适合线性、非线性,稳定、非稳定信号的分析。它运用经验模态分解方法筛选出一系列的本征模态函数和最后的残余量,具有良好的时频局部性,被广泛应用于一维信号分析和二维图像处理研究领域。本文主要介绍了基于经验模态分解EMD方法的医学图像融合方法,提出了基于EMD-PCA的医学图像融合新算法。并与其他融合方法进行对比实验,证明了本算法对于医学图像融合的可行性和优越性。更多还原

【Abstract】 Image fusion is a comprehensive data processing, which used two or more images that are obtained from acquisition devices. to combine information from multiple images more clear, the amount of information larger, higher quality of fusion images. Medical image technology is a modern medical treatment method, which utilized all kinds of imaging devices to observe the internal organization, bone structure and functional change of the human body. Different medical imaging devices provide different medical information, and play a different role in medical imaging and clinical diagnosis. Medical image fusion can be well integrated information from multiple images, and improve the accuracy of clinical diagnosis and treatment.In 1998. Norden E. Huang from NASA. Chinese-American proposed a new time-frequency analysis theory:Hilbert-Huang transform (HHT). For existing signal analysis as Fourier, or wavelet transform depending on the predetermined basis functions, does not have the data adaptive. HHT transform has fully adaptive, widely used in seismic studies, marine, voice, and mechanical failure analysis field. It used empirical mode decomposition (EMD) to decompose a complex signal into a series of intrinsic mode function (IMF) combined. It is suitable for non-linear and non-stable signals. with good time-frequency characteristic based on local characteristics of the signal.With the EMD theory has mature applications in one-dimensional signal processing, domestic and foreign scholars extend it to two-dimensional image processing, called the Two-dimensional empirical mode decomposition (BEMD). For the sifting of the BEMD process, there are four core steps:First, search the maxima and minima extreme points of the image. Second, choose the envelope surface interpolation function. This needs to consider surface smoothness and computational speed, also need to tolerate the irregular data points, and can be good to avoid the overshoot phenomenon. Third, research the stopping criteria for sifting and decomposition. Fourth, solve the border problem. When we do the decomposition process, if data dispersal at the end of the lower and upper envelopes, it will dispersal from the end to the inner gradually, and ’pollution’ the whole data with the sifting running, resulting in decomposition serious distortion.In this papers, consider the medical images’ own characteristics, I proposed nearest 8 neighborhood searching method to extract the extreme points, interpolation method based on Delaunay triangulation.’SD’values and fixed sifting number as the stopping criteria, and the even extension to solve the image boundary effect as the 4 key BEMD sifting algorithm to solve the above problems. It is also in line with the needs of medical image fusion.The prime task of this paper is to do image fusion research based on BEMD method. We do the BEMD image fusion experiments with the weighting coeffient fusion rules. On this basis, proposed a new medical image fusion algorithm, it is based on empirical mode decomposition and principal component analysis (EMD-PCA) method. This method using principal component analysis on the intrinsic mode functions after EMD decomposition. to calculate the eigenvalues and eigenvectors, then according to the proportion degree to data fusion. Also medical image fusion evaluation criteria were introduced, and compare the final image with the traditional image fusion method for the quantitative and qualitative comparison. The series of evaluation proved that the algorithm used in this paper enhances spectral information and details, showed that the new fusion method is better.At the last part of the paper, the article summarizes the results of this study, pointed out that EMD is a good time and frequency signal analysis and processing tool. Also analyzes the deficiencies of the EMD theory, and finally predicts the trends and directions of EMD theory.更多还原