摘要
为实现视网膜血管在临床诊断中的重要作用,提出了一种基于Hessian矩阵的视网膜血管提取方法,该方法通过图像预处理增强血管信息,利用血管的微分几何特性,采用离散高斯核对眼底图像进行卷积,结合Hessian矩阵计算血管方向,通过连接算法得到视网膜血管的分布情况。实验结果表明,该方法提取血管中心线的精度可达亚像素级,对不同眼底照相机拍摄的眼底图像可根据血管宽度进行多尺度快速分割。
The detection and segmentation of retinal blood vessels have great significance in the diagnosis of cardiovascular and cerebrovascular diseases.It proposes a method of the extraction of retinal blood vessel centerlines based on Hessian matrix.The pre-processing of fundus images enhances the vessels and convolute the image with the masks of discrete Gaussian partial derivative kernels.The directions of blood vessels are determined by Hessian matrix.according to the differential geometric properties of the vessels,the vessel centerline points are calculated and connected.The experiments on several kinds of fundus images demonstrate the multi-scale extraction according to different widths of retinal blood vessels.The precision of this method can reach sub-pixel level with a fast computation speed.
