【作者】 刘莉莉；

【导师】 王铮；

【作者基本信息】 重庆大学 ， 计算机软件与理论， 2009， 硕士

【摘要】 类圆性颗粒图像的处理与分析在工农业生产、医疗卫生等领域均有广泛的应用。将这些物质颗粒从图像中分割出来,统计其数目并提取各单个物质颗粒的特征参数(如面积、周长、直径、中心矩、颜色等),可以对其品质进行分析。例如,确定菌落数量是农业、食品、医药卫生分析中进行质量检测的一项基本而重要的工作,水中的菌落计数是评价水受污染程度的一项重要质量指标。早先,这些工作主要由人工观测来完成,工序繁杂、耗时长、效率低,而且带有一定的主观性,误差大,重现性不好。采用图像处理与分析的方法能够将操作人员从这一繁重的工作中解脱出来,并大大提高计数与分析的精度,从而得到广泛的应用,成为近年来国内外的一大研究热点。图像分割是颗粒图像处理与分析过程中的关键环节,分割结果的好坏直接影响计数的精度与后续的处理。然而,图像分割技术也是图像信息工程中的一大经典难题,尽管众多国内外学者对此进行了广泛、深入的研究,提出了不少应用算法,但仍没有一种方法对所有测试图像分割效果均为最佳。其中细胞的自动识别在计算机辅助诊断中,尤其是缺少专家的情况下具有十分重要的意义。由于细胞图像中细胞尺寸可能大小不一、形态各异,并且往往产生聚堆现象,分割的难度更大。目前还没有一种完全通用的分割方法可以实现对所有细胞图像的正确分割。本文主要对粘连细胞这一细胞分割领域中的难点问题进行研究。在研究传统预处理技术及分割算法的基础上,分析他们的优点和不足。提出一种基于局部流域变换的迭代剥离算法。具体分割过程分为两个阶段。首先,在预处理领域引入数学形态学,采用和传统预处理技术相结合的处理方法,有效地消除了背景噪声和空洞干扰,并且设定合理区域圆形度Pth ,标记并去除图像中的单个籽粒区域;在分割阶段采用改变了流程的流域分割,整个过程是基于局部流域变换的迭代剥离,消除过分割现象。血细胞图像计数与分析系统。采用软件工程开发方法,构建了一个图像处理与分析框架系统,包含丰富的图像处理与分析算法函数。更多还原

【Abstract】 The processing and analysis of circle-like particle image has a wide application in industry, agriculture, and medical treatment, such as the grain, rock, colony, blood cell, bubble, bar, and etc. After these particles being segmented from the background, we can count the number, extract their features (e.g. area, perimeter, diameter, color etc.), and then we can analyze its quality. For example, colony counting is a basic and important job for quality inspection in agriculture, food, and medical treatment analysis. The number of colonies is an important quality parameter in evaluating the water pollution. In earlier, it is inspected mainly by operator and the process is complicated, time-consuming, and low-efficiency. Moreover, the result tends to subjective, error distinctive, and bad recurring. Image processing and analysis can free operator from the labored work, and can improve the counting and analysis accuracy highly. As a result, it is widely applied and becomes a research hotspot in recent years.Segmentation is a key step during the particle image processing. The segmentation result has direct influence on the counting problem in image information engineering. Though numerous domestic and international scholars have extensive lucubrated into this subject and have presented many applicable algorithms, there is still no such an algorithm that can be applied to any test image and get the best result.And the cell auto-recognition systems are significant in computer assistant diagnosis, especially in the instances of lacking experts. Since the cell images are various in size and modality, and tend to cluster into each other, the cells are more difficult to be segmented, by now there isn’t a segmentation algorithm that fits all kind of images. In this paper, the problem of the image segmentation of overlapping cells is mainly studied. First, traditional pretreatment technologies and segmentation methods are discussed. We can know merit and shortcoming of them. And an improved algorithm based on local watershed transform was presented in this work. It is constructed with two similar stages. Firstly, the mathematics morphology is produced. The traditional pretreatment technologies and mathematics morphology can eliminate background noise and holes effectively together. And single seed regions are marked and removed from the clustering paddy image according to circle threshold( Pth ); Adopt different process watershed algorithm, Since the whole process is based on local watershed algorithm, it eliminate the phenomena of over-segmentation. Cells counting and analysis system developing. With software engineering developing methods, an image processing and analysis frame system is constructed which covers rich image processing and analysis algorithms.更多还原