【作者】 马睿；

【导师】 曾理；

【作者基本信息】 重庆大学 ， 应用数学， 2008， 硕士

【摘要】 随着科学技术和工业生产的迅猛发展,许多领域越来越多地提出了对三维工件尺寸和形状参数测量的要求。近年来,利用CT进行检测时,从定性检测向定量测量发展是其中一个重要的研究方向。而CT图像测量就是以CT断层图像为研究对象来获得目标几何参数的测量方法,目前已广泛应用于工业检测和医学诊断等领域。随着实际应用对测量精度要求的不断提高,已有的测量方法很难满足实际需要,因此高精度测量方法的研究具有重要的应用价值。本文完成的主要研究内容如下:在图像测量领域,待测目标边缘点的定位精度往往直接影响到整个测量的精度。传统的像素级边缘检测方法精度不高,因此讨论了亚像素边缘检测技术,尤其重点研究基于Facet模型的亚像素边缘检测算法,该算法具有定位精度高、抗噪能力强等优点,但时间复杂度太高。针对此缺点,提出了一种改进算法,先采用Mallat的小波变换模极大算法提取初始边缘点,再用2D Facet模型对初始边缘点进行处理,进一步确定边缘点,并计算其亚像素精确位置。改进算法将基于2D Facet模型的算法与小波变换模极大算法有效地结合起来,弥补了各自的缺点,发挥了它们的优点,不仅提高了处理速度,而且增加了抗噪能力和边缘的连续性。针对常用二维测量方法精度不高等不足,研究了一种亚像素二维测量方法,并将其应用于实际的工业CT图像测量中,提高了二维测量的精度。首先,采用改进算法提取亚像素边缘,为测量提供了高精度的数据;然后,通过最小距离搜索法分离出待测目标的边缘点并生成排序链码,克服了边缘点是浮点型且不连续的困难,给参数的计算提供了有效的数据;最后,分别利用离散化的格林公式和欧氏距离公式计算目标的面积和周长。针对常用三维测量方法精度不高、自动化程度低等不足,研究了一种亚体素三维测量方法,并将其应用于实际的工业CT体数据测量中,提高了三维测量的精度和自动化程度。首先,采用基于3D Facet模型的亚体素边缘检测算法提取边缘,该算法充分利用了相邻层的相关信息,能提取更准确的边缘;然后,通过降维处理和最小距离搜索法把待测目标的边缘点从整体边缘中分离出来并生成排序链码;最后,利用最短对角线法用很多基本三角面片重构三维目标的表面,由形心向每个三角形的每个顶点连线得到很多三棱锥,分别累加三棱锥体积和三角形面积计算目标的体积和表面积。本文对工业CT图像测量做了比较系统的研究,在分析已有方法不足的基础上,建立了高精度的测量方法,并通过对比实验对测量的精度进行了验证。与已有方法不同的是,本文的测量方法都是在亚像(体)素边缘上进行的,突破了图像分辨率对测量精度的限制,使在中低分辨率的图像上实现高精度的测量成为可能。更多还原

【Abstract】 As rapid development of technology and industrial production, the geometrical parameters measurement of 3D object is proposed by many fields. Recently, when the CT is used, the transfer from test to measurement is also a main research direction. CT image measurement is a method that measures the parameter through processing CT slices, and it is used in many fields, such as industrial testing and medicine examining. Applications ask for higher and higher accuracy of measurement, but the accuracy of widely used methods are not high enough. So the measurement method of higher accuracy must be proposed by request. The main research contents and contribution in the thesis are as follows:In the applications such as image measurement, the object’s edge information of high accuracy is required. First, a sub-pixel edge detection method based on Facet model is introduced, the method can reduce noise and achieve high accuracy, but its computational complexity is too high. Aimed at making up this disadvantage, we studied an improved method, which combined the 2D Facet model and Mallat’s maximum wavelet module approach effectively. The wavelet method is used to extract wide preparatory edge, while restrain some noise. Then the method based on 2D Facet model merely processes preparatory edge points and further obtains sub-pixel edge. The improved method not only increases the speed, but also provides more continue accurate edge and reduces noise.In order to improve the accuracy of 2D image measurement, a sub-pixel measurement method was studied, and applied in actual industrial CT images. Firstly, the improved sub-pixel edge detection method is used to extract the sub-pixel edges. This method can achieve higher accuracy, reduce noise, and offer accurate data for further area measurement. Secondly, the method of minimizing distance search is studied to separate and sort the edge of the measured object. This search method can separate the edge points of the measured object from the floating-point and discontinuous edges of whole image, obtain sorted edge points chain, and offer available data for the next computing. At last, Green formula and Euclid distance formula are adopted to compute the 2D parameters.In order to improve the accuracy and automatism of 3D measurement, a sub-voxel measurement method was studied, and applied in actual industrial CT voxel data. Firstly, an edge detection method based on 3D Facet model is used to extract the sub-voxel edges. This method considers the information of slices adjacent to the being processed one, can reduce more noises and achieve higher accuracy when compared with the 2D sub-pixel edge detectors which only use single slice information. Secondly, the method of reducing dimension and minimizing distance search are studied to separate and sort the edge of the measured object. At last, the surface of the object is reconstructed by many small triangles using shortest diagonal method, and the three 3D can be compute through these triangles.This thesis has studied on industrial CT image measurement. Based on anlysising disadvantages of prevenient measurement methods, we present the measurement methods of higher accuracy. The experimentations have done on emulational images and actual industrial CT images, and the results show the validity of the method. The measurement methods in this paper are based on sub-pixel (sub-voxel) edge, which breaks through the restriction of measurement precision by image resolution.更多还原