医学图像分割与病变特征提取研究

发布时间：2018-03-11 13:52

本文选题：图像分割　切入点：Otsu方法　出处：《太原科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：图像分割在数字图像处理、模式识别等领域是非常重要的研究课题,尤其在医学领域中发挥着越来越大的作用。由于传统的医学图像分割基本上是基于人工分割,分割结果往往无法令人满意,而且费时费力。因此,如何将医学图像实现自动分割,并以此为基础对病灶图像进行识别等,一直是医学图像处理的研究难点和重点。本文针对未考虑灰度信息分布不均匀而使得一些细节未能得到良好分割的情况,将主动轮廓法和最大类间方差Otsu方法相结合来进行分割;在此基础上,考虑病变区域纹理特征和形状特征信息,通过支持向量机SVM对其病灶图像进行了分类识别处理,主要内容如下:(1)医学图像区域分割研究。针对各种具有复杂的器官与组织的医学图像中常常未考虑到灰度信息分布不均匀的情况,本文将Otsu方法融入到水平集Chan-Vese模型中,构造新的能量函数,对分割图像进行目标轮廓演化处理,在保留了Chan-Vese模型优点的情况下,融入了图像分布的类间方差信息,从而实现灰度信息分布不均匀医学图像分割。采用两个数据集提供的人脑图像数据进行实验,结果表明所提方法在相似性度量和正误率度量方面,相比其他同类方法都有明显的优势。(2)病变图像特征提取与识别研究。针对医学图像病变区域信息复杂,使用单一的纹理特征分类效果不佳的问题,本文首先在常用的纹理特征基础上,融入Hough变换和不变矩两个形状特征,以考虑旋转、平移等畸变带来的图像失真影响;然后对这些非线性分布融合信息线性化处理,通过支持向量机SVM对其进行分类,得到病变图像与正常图像的识别;最后,采用某医院提供的图像进行纹理特征和形状特征提取,进而进行SVM分类实验,实验结果表明,分类准确率有所提高。
[Abstract]:Image segmentation is a very important research topic in the fields of digital image processing and pattern recognition, especially in the field of medicine. The segmentation results are often unsatisfactory and time-consuming. Therefore, how to segment the medical image automatically and recognize the focus image based on it, etc. It has always been a difficult and important point in medical image processing. In this paper, some details are not well segmented because the uneven distribution of gray information is not considered. The active contour method and the maximum inter-class variance (Otsu) method are combined to segment the lesions. On this basis, considering the texture and shape features of the lesion region, the focus images are classified and identified by support vector machine (SVM). The main contents are as follows: (1) Research on region segmentation of medical image. In view of the non-uniform distribution of gray information in various medical images with complex organs and tissues, the Otsu method is incorporated into the level set Chan-Vese model. A new energy function is constructed to process the target contour evolution of the segmented image. With the advantages of the Chan-Vese model, the inter-class variance information of the image distribution is incorporated. In order to realize the segmentation of medical image with uneven distribution of gray information, the human brain image data provided by two data sets are used to carry out experiments. The results show that the proposed method is in the aspect of similarity measurement and correct and false rate measurement. Compared with other similar methods, it has obvious advantages in feature extraction and recognition of pathological image. Aiming at the problem of complex information of lesion region in medical image, the classification effect of single texture feature is not good. In this paper, based on the commonly used texture features, Hough transform and moment invariant feature are incorporated to consider the distortion effect caused by the distortion such as rotation and translation, and then the nonlinear distribution fusion information is linearized. It is classified by support vector machine (SVM) to obtain the recognition of pathological image and normal image. Finally, the texture feature and shape feature are extracted from the image provided by a hospital, and then the experiment of SVM classification is carried out. The experimental results show that, The accuracy of classification is improved.
【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R310;TP391.41

【参考文献】