基于Hessian矩阵和水平集的视网膜血管分割  

Retinal Blood Vessel Segmentation Based on Hessian Matrix and Level Set

作　　者：梁礼明[1] 黄朝林[1] 朱莎[1] 吴健[1] 

机构地区：[1]江西理工大学电气工程与自动化学院,赣州341000

出　　处：《科学技术与工程》2016年第10期44-49,共6页Science Technology and Engineering

基　　金：国家自然科学基金资助项目(51365017);江西省自然科学基金(20132BAB203020);江西省教育厅科学技术研究(GJJ13430)资助

摘　　要：视网膜血管图像可以作为糖尿病与青光眼等疾病的诊断依据;但现有的血管提取算法存在精度不足、对噪声敏感,以及鲁棒性不强等缺点。针对这个问题提出了一种新的视网膜血管图像分割方法。首先获取视网膜绿色通道图像,利用Hessian矩阵特征值、特征向量的几何特性和响应函数,初步估计视网膜图像中可能存在的血管;并在此基础上利用水平集函数初始化血管图像。然后,在局部数据能量拟合泛函中引入新的正则化和面积约束。最后,在水平集函数演化过程中获得精准的视网膜血管图像。实验表明算法在分割视网膜血管图像上具有较强的鲁棒性。The image of retinal blood vessel is often used as a basis of the diagnosis for diabetes and glaucoma.However,existing algorithms have the disadvantages of low accuracy,sensitivity to noise,and bad robustness.For this problem,retinal vessel extraction was proposed method based on Hessian matrix and level-set.First,the green channel of retinal image is obtained.Potential blood vessels of retinal images are preliminary extracted via Hessian matrix eigenvalues,feature vectors of the geometric features and the response function.Then,a new regularization and area constraint is introduced into the local data energy fitting functional.Finally,the precision of retinal blood vessel image is obtained in the evolution of the level set function.Experiments show that the proposed algorithm has better performance to the segmentation of retinal blood vessels.

关 键 词：HESSIAN矩阵 响应函数 水平集 面积约束 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]