基于支持向量机的血细胞图像分类识别技术的研究-计算机软件与理论专业论文.docxVIP

HYPERLINK \l _bookmark0 Abs HYPERLINK \l _bookmark0 Abstract II万方数据 II 万方数据 Abstract The resarch on blood cells image is an important branch in the field of medicine image processing. It mainly includes following three sections: segmentation of cells region, segmentation of overlapping cells and classification of leukocyte. Through the comparision of the previous research, this paper introduces a new method of processing blood cells image based on SVM, which has a good capability on classification and generalization. The main contents and contributions are as follows: Based on the theroy of that the leukocyte segmentation problem could be converted into the problem of pixels classification, a method of segmention of image of blood cells based on one-against-others SVM is given. The effects of input vectors, kernel function and other parameters are investigated by experiment. Porting of LIBSVM2.83 is also discussed. With regard to the segmentation of clustering cells, an improved watershed segmentation algorithm based on distance transformation is presented on the basisi of previous research, which uses a new method of distance transformation by ultra-erosion To record the real space location relationship of every pixel with a better result. By analysing the characteristics of various kinds of leukocyte, followed by extracting 14 characteristics f shape, texture, and colore distilled from the area of leukocyte kernel into SVM input space, this paper presents a five-classification based on one-against-one SVM. Because of limited leukocyte image samples, its experiments are elementary. Keywords: SVM; segmentation of blood cells image; segmentation of clustering cells; classification of leukocyte; LIBSVM2.83 HYPERLINK \l _bookmark1 目 HYPERLINK \l _bookmark1 目 录 PAGE IV PAGE IV 万方数据 目 录 摘 要 I Abstract II 目 录 III 第一章 绪论 1 §1.1 课题的研究意义 1 §1.2 国内外研究现状 2 §1.3 本文的研究内容 3 §1.4 本文的内容安排 3 第二章 统计学理论与支持向量机 5 §2.1 机器学习 6 §2.1.1 机器学习问题的描述 6 §2.1.2 经验风险最小化原则 6 §2.1.3 模型复杂性与泛化