基于KPCA与SVM的视网膜微动脉瘤检测算法的研究

发布时间：2018-01-20 19:10

本文关键词： 视网膜微动脉瘤主成分分析核主成分分析支持向量机　出处：《东北大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着计算机技术的发展,数字视网膜图像处理与分析技术也日渐成熟。糖尿病性视网膜病变是一种严重的糖尿病并发症,是导致新发失明的主要原因。目前,在糖尿病的诊疗过程中面临的重大难题是如何安全高效地进行数字视网膜图像筛查,及时地识别糖尿病性视网膜病变以便尽早地采取相应措施避免失明。本文主要针对糖尿病性视网膜病变中最早出现的微动脉瘤病征进行研究,提出了一种基于核主成分分析(KPCA)和支持向量机(SVM)的微动脉瘤检测算法。根据微动脉瘤的检测原理可知,视网膜微动脉瘤的检测过程主要分为三大部分：微动脉瘤的候选集获取、微动脉瘤特征抽取以及微动脉瘤特征分类。本文基于图像预处理技术和微动脉瘤的粗提取理论获取了视网膜微动脉瘤候选集,并深入研究了视网膜微动脉瘤的特征抽取算法和特征分类算法。本文首先研究了基于PCA的微动脉瘤特征抽取算法,该算法的实质是通过线性映射将微动脉瘤的特征样本从高维的特征空间变换到较低维的特征子空间的过程。然而,由于PCA算法本身的局限性,在进行变换的过程中忽略了高阶特征指标之间的相互关系,使降维后的主成分并不能很好地表达原来的高维特征指标所携带的信息。为解决该不足,本文提出了基于KPCA的微动脉瘤特征抽取算法,该算法首先通过核函数将高维特征向量映射到核空间,然后再在核空间上进行PCA处理,以实现高效地特征抽取。另外,本文在主成分分析的基础上借助于截断误差法设计相应的分类器,实现了对低维微动脉瘤候选集的分类。实验证明,虽然两种算法均能达到特征抽取的目的,但是基于KPCA的特征抽取算法更能有效地降低特征维数,提高微动脉瘤的检测真阳性率。其次,本文设计使用基于SVM的特征分类算法对特征抽取后的低维微动脉瘤候选集进行分类。针对传统算法中单纯采用各个特征量区间约束策略时忽视各个特征量的约束力大小的弊端,本文采用了基于统计学理论基础的SVM分类算法,算法中关于核函数参数与惩罚因子的确定,本文采用粒子群优化算法选取最优的参数值。实验证明,基于SVM特征分类算法能够有效的降低微动脉瘤的检测假阳性率。实验结果显示,综合使用KPCA和SVM算法能有效提高微动脉瘤检测算法的检测精度。
[Abstract]:With the development of computer technology, digital retinal image processing and analysis technology is becoming more and more mature. Diabetic retinopathy is a serious complication of diabetes, which is the main cause of new blindness. In the course of diagnosis and treatment of diabetes mellitus, the major problem is how to screen digital retinal image safely and efficiently. In order to identify diabetic retinopathy in time and take appropriate measures to avoid blindness as early as possible, this paper mainly focuses on the microaneurysm symptoms of diabetic retinopathy. A microaneurysm detection algorithm based on kernel principal component analysis (KPCA) and support vector machine (SVM) was proposed. The detection of retinal microaneurysms is mainly divided into three parts: the acquisition of candidate sets of microaneurysms. Microaneurysm feature extraction and microaneurysm classification. Based on image preprocessing technique and rough extraction theory of microaneurysm, the candidate set of retinal microaneurysm was obtained in this paper. The feature extraction algorithm and feature classification algorithm of retinal microaneurysms are studied in depth. Firstly, the feature extraction algorithm of microaneurysms based on PCA is studied in this paper. The essence of the algorithm is to transform the feature samples of microaneurysms from the high-dimensional feature space to the lower-dimensional feature subspace by linear mapping. However, due to the limitations of the PCA algorithm itself. In the process of transformation, the relationship between high-order feature indexes is neglected, so that the reduced principal component can not express the information carried by the original high-dimensional feature index. In this paper, a feature extraction algorithm for microaneurysms based on KPCA is proposed. Firstly, high dimensional feature vectors are mapped to kernel space by kernel function, then PCA is processed in kernel space. In order to achieve efficient feature extraction. In addition, based on principal component analysis (PCA), the corresponding classifier is designed based on the truncation error method to realize the classification of low-dimensional microaneurysm candidate sets. Although the two algorithms can achieve the purpose of feature extraction, the feature extraction algorithm based on KPCA can effectively reduce the feature dimension and improve the true positive rate of microaneurysm detection. In this paper, the feature classification algorithm based on SVM is designed to classify the low-dimensional microaneurysm candidate set after feature extraction. The drawbacks of binding size. In this paper, SVM classification algorithm based on statistics theory is adopted. In the algorithm, the parameters of kernel function and penalty factor are determined, and the particle swarm optimization algorithm is used to select the optimal parameter value. Based on SVM feature classification algorithm can effectively reduce the false positive rate of microaneurysm detection. Experimental results show that the combined use of KPCA and SVM algorithm can effectively improve the detection accuracy of microaneurysm detection algorithm.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R587.2;R774.1;TP391.41

【相似文献】