基于HESSIAN增强和形态学尺度空间的视网膜血管分割

RETINAL VESSELS SEGMENTATION BASED ON HESSIAN ENHANCEMENT AND MORPHOLOGICAL SCALE SPACE

眼底视网膜血管的走向、弯曲度、分叉度等性状分析已成为医学上诊断全身血管性疾病的重要手段。采集到的眼底图像常存在光照不均匀等现象,利用传统的血管分割方法难以对微小血管进行检测。为此提出一种基于改进Hessian矩阵增强和形态学尺度空间的分割方法。首先利用高斯函数构建多尺度Hessian增强滤波器,采用新型的血管相似性函数对血管网络进行对比度增强,同时平滑图像以减轻噪声;然后利用改进的Top-hat变换尺度空间从背景中提取血管,并引入形态学重建方法进一步突出血管像素,消除伪边缘及孤立点噪声;最后使用二次阈值化方法实现血管的最终分割。仿真结果表明,改进的分割方法在保证大血管脉络准确分割的同时,能够较好地实现微小血管分割。

Characters analysis in regard to the tre n d , curvature and b ifu rc a tio n o f re tin a l vessels in fundus has become the im po rtan t meanso f system ic vascular diseases diagnosis in m e dicine science. Because o f m ost collecte d fundus images has the phenom enon o f lig h tunevenness, it is d iffic u lt to use tra d itio n a l vessel segm entation methods to detect the m icro vessels. Therefore we proposed a segm entationa lg o rith m , it is based on the im proved Hessian m a trix enhancem ent and m orphological scale space. F ir s t, by using Gauss fu n c tio n thea lg o rith m constructs m u lti-sca le Hessian enhanced f ilt e r , and uses a novel vascular s im ila rity fu n c tio n to carry out the contrast enhancem ent onvascular n e tw o rk , w h ile smoothes the image to w eaken noise as w e ll; then it extracts the vessels fro m background using an im proved T o p-h attransform ation scale space, and introduces m orphological re construction m ethod to fu rth e r h ig h lig h t the vascular pixels and to e lim in a te thepseudo-edges and the noise o f o u tlie rs; fin a lly the a lgo rithm uses secondary thresho ld ing approach to realise fin a l vessel segm entation.S im u la tio n experim ental results showed that w h ile ensuring the accurate segm entation o f great vessels and c h o ro id , the im proved segm entationm ethod can be tter realise the segm entation o f m icro vessels.