基于稀疏表示和字典训练的微血管分割方法研究

发布时间：2018-08-01 08:33

【摘要】：微血管的状态信息与人体组织、器官代谢水平具有密切的关系,当微血管中血液流动的状态出现异常时,可以合理的推断出机体的某个部位发生了病变,因此,微血管信息具有重要的生理、病理、药理和临床意义。对微血管的识别研究也在各类疾病的早期诊断以及治疗方面起着不可小觑的作用。同时,随着计算机的高速发展,利用计算机来完成对数字图像信息的处理和分析工作已变得十分常见,通过计算机对微血管数字图像进行识别,不仅可以实现对生物体的零创伤,同时还可以加快数据的处理速度,提高了处理效率,减轻了科研人员的压力。因此,利用数字图像处理技术完成对微血管的识别研究,在生物科学、医学诊断等方面有着十分重要的意义。本文从微血管图像的特征以及图像分割两方面入手,首先研究了微血管图像的特征以及常见的图像分割方法,包括基于阈值分割,边缘检测,形态学运算以及区域生长四种方法,并实现了对金鱼尾部微血管的分割。其次,本文学习了稀疏表示和字典学习理论。稀疏表示理论的应用领域非常广泛,如图像压缩,图像去噪以及图像分割等,且都得到了较为优质的效果。因此本文在稀疏表示和字典学习理论的基础上,学习了基于稀疏聚类的图像分割方法,并建立了基于稀疏子空间聚类的图像分割模型。该模型运用Ncut法对图像划分出N个超像素,然后应用SAC算法计算每个超像素之间的相似性,并利用相似性矩阵求得系数矩阵A,运用系数矩阵A构造图像的邻接矩阵W,最后再次运用Ncut方法对超像素进行划分从而得到图像的分割结果。本文的最后以活体的非洲爪蛙作为实验材料通过一系列的实验完成了对微血管图像的采集,灰度变换,融合以及分割的处理。其中包括,运用灰度变换法将采集到的图像转换成灰度图像,方便后续计算机的进一步处理;运用像素点选小融合法对转换后的微血管灰度图像进行融合处理,将所采集到的不连续的微血管图像融合成一幅完整的微血管显微图像,从融合后的结果中可以清晰地看到一条连续的微血管脉络;运用基于块稀疏子空间聚类模型对图像进行分割处理,并将分割结果与前文提到的四种常用图像分割方法做比较,可以明显的看出,本文所使用的方法可以将微血管更加完整清晰的分割出来,分割效果更好。
[Abstract]:The state information of microvessels is closely related to the metabolism level of human tissues and organs. When the blood flow in the microvessels is abnormal, it can reasonably be inferred that there is a pathological change in a certain part of the body. Microvascular information has important physiological, pathological, pharmacological and clinical significance. The recognition of microvessels also plays an important role in the early diagnosis and treatment of various diseases. At the same time, with the rapid development of computer, it has become very common to use computer to process and analyze digital image information. It can not only achieve zero trauma to organism, but also speed up data processing, improve processing efficiency and lighten the pressure of researchers. Therefore, the recognition of microvessels by digital image processing technology is of great significance in biological science, medical diagnosis and so on. This paper starts with the features of microvascular image and image segmentation. Firstly, it studies the features of microvascular image and common image segmentation methods, including threshold segmentation, edge detection. Morphological operation and regional growth were used to segment the tail microvessels of goldfish. Secondly, this paper studies sparse representation and dictionary learning theory. Sparse representation theory is widely used in many fields, such as image compression, image denoising and image segmentation. Therefore, based on sparse representation and dictionary learning theory, this paper studies the image segmentation method based on sparse clustering, and establishes an image segmentation model based on sparse subspace clustering. The model uses the Ncut method to divide the image into N superpixels, and then uses the SAC algorithm to calculate the similarity between each superpixel. The coefficient matrix A is obtained by using the similarity matrix, and the adjacent matrix W is constructed by using the coefficient matrix A. finally, the image segmentation results are obtained by using Ncut method to partition the super-pixels. In the last part of this paper, a series of experiments were carried out to collect, transform, fuse and segment the microvascular image with the living African claw frog as the experimental material. Among them, the grayscale transform method is used to convert the collected image into gray image, which is convenient for the further processing of the subsequent computer; the image of microvascular grayscale after conversion is fused by the method of pixel selection and small fusion. The collected discontinuous microvascular images are fused into a complete microvascular image, and a continuous microvascular vein can be clearly seen from the fusion results. Using the block sparse subspace clustering model to segment the image, and comparing the segmentation results with the four commonly used image segmentation methods mentioned above, we can clearly see that, The method used in this paper can segment the microvessels more completely and clearly, and the segmentation effect is better.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R445;TP391.41

【相似文献】