【作者】 孟现海；

【导师】 刘以安；

【作者基本信息】 江南大学 ， 计算机软件与理论， 2008， 硕士

【摘要】 图像边缘检测和图像配准是图像融合的关键步骤和必要前提。本文对图像的边缘检测和图像配准技术进行研究,把一些新的算法运用到图像边缘检测和图像配准中,为后续序列图像处理和图像融合做准备。边缘特征是图像中非常重要且容易获得的特征,已经有很多边缘提取的算法,例如sobel算子、canny算子、Log算子等。但这些算法对噪声比较敏感,虽然改进的canny算子有了很大的提高,能提取出比较清晰的边缘,并具有一定的抗噪性,但是算法检测速度较慢,不能用于序列图像处理中。为了寻找具有检测速度快、抗噪性强、检测精度高以及边缘细节保护好的检测算法,本文把集对分析和联系度态势的思想用到图像的边缘检测中。先用集对的方法求出像素点八个方向的同一度、对立度和差异度,再用联系度态势的思想把像素点的同异反关系按同势、均势、反势的趋势进行排序,然后根据像素点的趋势关系来判别该点是否是边缘点。另外有些图像不仅对比度差,而且图像的边缘轮廓也较模糊,所以在进行图像的边缘提取之前可以先对图像进行灰度变换,增加图像的对比度和突出图像的边缘特征。仿真结果表明此算法不仅得到了较好的边缘,而且算法的检测速度也较快。图像配准的方法有很多种,其中基于图像特征的图像配准是配准中最常见的方法。基于特征的图像配准中,特征主要针对点特征。为了得到一种配准速度较快的高配准率算法,本文用的是基于特征点的配准方法,即先用SUSAN算子来提取图像的特征点,再用PSO算法在解空间内搜索最佳匹配参数,然后进行图像的配准。在SUSAN算子中,灰度差阀值t决定了SUSAN算子所能检测到的最小的对比度以及去除噪声点的能力,本文对t值进行了改进,给出了一种对t值自适应的提取方法。PSO是一种新的并行优化算法,可以解决大量非线性、不可微、非连续性和多峰的复杂问题,但是该算法易陷入局部最优,会出现所谓的早收敛现象。为了克服PSO算法的缺点,提出了将Alopex算法加入到PSO算法的改进算法,这样有利于PSO算法在搜索中跳出局部极值,同时又能根据目标函数的变化加速算法的收敛。最后用一幅红外图像、微波图像和多光谱图像作为实例来验证此算法,分别在算法的迭代步数、时间和准确度方面与ICP算法和改进前的PSO算法作比较,通过实验结果可以看出,本文实现的配准方法能对图像进行有效的配准。更多还原

【Abstract】 Image Edge detection and image registration is a key step and the necessary prerequisite of image fusion. Image edge detection and image registration techniques are studied and some new algorithms are applied to image edge detection and image registration in this paper, to prepare for sequence of image processing and image fusion.In the image, edge characteristics is very important and easy acquired characteristics, there are many of edge detection algorithms, such as Sobel operator, Canny operator, Log operator are being applied to image edge detection. But these algorithms are very sensitive to noise, although improved Canny operator has greatly increased in anti-noise and relatively clear edge that can be extracted, but the detection speed of the algorithm is slower, so that can not be used in the image processing sequence. To find a testing speed, strong anti-noise, high detectable precision and better edge-details protection algorithm, set pair analysis and degree connection situation are applied to the image edge detection in this paper. firstly, the degree of identity ,opposition and discrepancy of the eight directions of the pixel are calculated with set pair analysis, secondly, the IDC (Identical-Discrepancy-Contrary) connection of the pixel is ordered in order to Identical-Balanceable-Contrary trend with degree connection situation, and then determine whether the point is the edge by trend of the pixel. In the others, contrast is not only poor, but also characteristic of image edge is fuzzy to some images, in order to increase the image contrast and highlight characteristic of image edge, gray-scale transformation of the image is operated before image edge detection. Our experiment indicated that the method has not only better distinct edge but also has faster speed.There are many methods that based on image characteristics is the most common method in image registration. Based on the characteristics of the image registration, the characteristic features is mainly on point. In order to obtain a registration faster high registration rate algorithm, the feature points Registration is applied in this paper. Firstly, image feature points are extracted by using SUSAN operator, secondly, the best matching parameters are searched in the search space by using of the PSO, and finally, the image registration is operated. In SUSAN operator, value of the t (that is gray level difference threshold) decide the smallest contrast and the ability to remove noise that SUSAN operator can detected, we improvement the value of the t , give a adaptive extraction methods that to the value of the t . PSO is a new parallel optimization algorithm, it can solve large nonlinear, non-differentiable, non-continuous and multi-complex problem, but the PSO easy fall into the local optimal, that is premature convergence phenomenon. In order to overcome the shortcomings of PSO algorithm, we propose that improving the PSO algorithm with Alopex the algorithm, it will be conducive to PSO algorithm jumping out of the local optimal in searching, at the same time accelerating the convergence of algorithm in accordance with the changes of the objective function. Finally, we verify the algorithm with an infrared image, a microwave image and a multi-spectral image, and compare with ICP and improving before the PSO algorithm in step, time and accuracy of the algorithm, the experimental results indicated that the method can effectively image registration.更多还原