基于主曲率和主方向的多尺度视网膜血管骨架提取方法  被引量:3

Multiscale retinal vascular skeleton extraction method based on principal curvature and principal direction

在线阅读下载全文

作  者:苑玮琦[1] 王安 Yuan Weiqi;Wang An(Computer Vision Group,Shenyang University of Technology,Shenyang 110870,China;College of Information Science and Electronic Technology,Jiamusi University,Jiamusi 154007,China)

机构地区:[1]沈阳工业大学视觉检测技术研究所,沈阳110870 [2]佳木斯大学信息电子技术学院,佳木斯154007

出  处:《仪器仪表学报》2021年第6期191-199,共9页Chinese Journal of Scientific Instrument

基  金:国家自然科学基金(61271365)项目资助。

摘  要:为了减少视网膜血管骨架提取过程中低对比度血管的漏检和误检数量,提出了一种基于主曲率和主方向的多尺度视网膜血管骨架提取方法。首先,分别提取视网膜图像中每个像素点在多尺度高斯滤波后的主曲率和主方向;其次,在每个尺度下分别提取最大主曲率方向上的局部最大值点,并通过曲率阈值筛选出高对比度血管中心像素点作为种子点;然后,利用最小主方向对低对比度血管进行骨架追踪和标记;最后,对多个尺度下提取的血管骨架进行融合。方法分别在DRIVE训练数据集、DRIVE测试数据集和STARE数据集上进行了测试,漏检数量分别为89、97、106,误检数量分别为99、101、122。实验结果表明,该方法能够提取出低对比度的细小血管骨架,但对于对比度在3个灰度级以内的细小血管存在少量漏检,对于与血管粘连的高对比度细条状纹理和病灶干扰存在少量误检。In order to reduce missed and false detection of low contrast vessels in retinal vascular skeleton extraction,a multi-scale retinal vascular skeleton extraction method based on principal curvature and principal direction is proposed.Firstly,the principal curvature and principal direction of each pixel in the retinal image is extracted after multi-scale Gaussian filtering.Secondly,the local maximum points in the direction of the maximum principal curvature are extracted at each scale,and the high contrast vascular center pixel is selected as the seed point by curvature threshold.Then,the minimum principal direction is used to track and labeled the low contrast vessels.Finally,the vascular skeleton extracted at multiple scales are fused.The proposed algorithm is evaluated on DRIVE training dataset,DRIVE test dataset and STARE dataset.The number of missed detection are 89,97,and 106,respectively.And the number of false detection are 99,101,and 122,respectively.Experimental results show that the proposed method can extract small vascular skeletons with low contrast,but there are a few miss detection for small blood vessels with contrast less than three gray levels,and a few false detection for fine stripe texture and lesions interference which adheres to blood vessels.

关 键 词:视网膜血管 主曲率 主方向 骨架提取 多尺度融合 

分 类 号:TP391.4[自动化与计算机技术—计算机应用技术] H79[自动化与计算机技术—计算机科学与技术]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象