【作者】 韩海梅；

【导师】 姚砺；

【作者基本信息】 东华大学 ， 计算机应用技术， 2011， 硕士

【摘要】 纤维图像分割是纤维图像自动识别系统中的一个重要步骤,是纤维特征提取以及纤维识别的基础。在纤维图像采集过程中,由于纤维包埋和切片技术、图像采集系统中的环境、设备及人为等因素的限制,采集到的图像存在光斑、扭曲变形、光照不均等噪声,这给纤维图像分割带来困难。本文在深入分析现有的图像分割方法的基础上,为解决传统图像分割方法用于纤维图像分割存在双边缘、伪边缘和轮廓不连续的问题,将Snake模型引入到纤维图像分割中,提出了一种基于Snake模型的纤维图像分割算法。本文在研究了Snake模型的基础上,为解决传统Snake模型对初始轮廓线位置敏感和不能分割凹陷区域等问题,重点研究了GVFSnake模型在纤维图像分割领域中的应用.虽然GVF Snake模型有效解决了传统Snake模型的缺点,但运用到纤维图像分割中也存在着很多难点,比如时间效率、初始轮廓线的有效区域等问题。针对这些难点,本文首先结合GVF Snake模型的特点和纤维图像自有的特性,研究了适用于GVF Snake模型的纤维图像平滑去噪、去除光照不均的预处理方法；其次,提出融合聚类分割算法和GVF Snake模型的纤维图像分割方法,选取K-means聚类分割结果作为GVF Snake模型的初始轮廓线,充分发挥二者的优点,有效地解决了传统纤维图像分割存在的诸多问题；最后,由于纤维图像本身的特点,上一步的分割结果中可能存在毛刺,本文在轮廓跟踪方法的基础上找到毛刺并将其去除,得到完整、连续且单像素的纤维分割结果。大量实验数据表明,利用本文算法分割纤维图像,有效解决了传统图像分割方法分割结果的双边缘、伪边缘或者轮廓不连续等问题,得到的结果准确、完整,实现了准确分割各类较复杂的纤维图像的目的,而且有效缩短了纤维图像的分割时间。更多还原

【Abstract】 Fiber image segmentation is an indispensable step in the automatic fiber recognition system, and it is the base of the processes of feature exaction, classification recognition. So to segment a complete, continuous and single pixel fiber is the critical task. With the limits of fiber embedding and sectioning technology and illuminated with point light source techniques in making fiber cross-sectional samples, the fiber images sometimes have problems such as distorting and uneven illumination. This situation brings fiber image segmentation many difficulties.Based on the research of classic image segmentation methods, for the limits of traditional segmentation algorithms such as false edge, double edge or discontinuous contour, Snake model is used in fiber image segmentation.The paper illuminates the foundation of Snake model, because Snake model cannot solve boundary concavities better and the original contour must be placed close to the real boundary of object, we introduce GVF (Gradient Vector Flow) Snake model. It can solve these problems better than Snake model. But there are also many difficulties in the fiber image segmentation, such as time efficiency or effective original contour. For these difficulties, the GVF Snake model is applied in the segmentation of fiber image, we analyzes and researches the pre-segmentation processing first, and then a new fiber image segmentation algorithm based on clustering segmentation and GVF Snake model is proposed. The clustering segmentation is used to obtain the original coarse contour of fiber; and next GVF Snake algorithm is applied to calculate the accurate fiber contour. At last due to the noise of fiber micrographic image, some fiber contours have burrs, which can be removed by contour tracing method.The experimental result shows that this algorithm is effectively and accurately, which can not only solve the problems such as false edge, double edge or discontinuous contour of the traditional segmentation algorithms, extract the complete and continuous fiber contour, but also depress the noise of fiber image.更多还原