视网膜眼底图像配准是临床眼科疾病诊断和治疗中的一个关键环节。针对眼底图像配准过程中大范围视场变化和过分割结构噪声等问题,该文提出了一种改进的基于图的视网膜图像血管匹配方法。将血管交叉点表示成图的顶点,把特征点间沿血管路径的相邻关系表示成边,进而在视网膜血管结构图中构造路径超边来刻画更高阶多元特征关系。在此基础上,实现了一种全自动的视网膜眼底图像配准算法。包括: 第一步,通过多尺度Gabor滤波算法来检测和提取视网膜血管网络; 第二步,利用一种高效的谱松弛匹配算法来求解两个路径超边图的顶点匹配对应关系。最后,通过特征匹配召回率统计和配准的血管中线距离误差两方面的实验,证明该文提出算法是有效和准确的。

Retinal fundus image registration is a key step in clinical eye disease diagnosis and treatment. Targeting the difficulties in dealing with pair of images with large field of view differences or with great structural noises caused by over-segmentation, this paper presents an improved graph-based algorithm to match retinal vessel networks by utilizing higher-order relations constructed from the vessel structural graphs whose nodes represent vascular bifurcations with the edges describing relations between feature points. The method performs in a fully automatic fashion, with a multi-scale Gabor filter first employed for detection and extraction of retinal vessels and with the correspondences then recovered between two pathwise hyper-edge graphs using an efficient pairwise spectral matching scheme. Experimental evaluation shows that the developed method is effective and accurate in terms of the feature recall rate and the vascular CEM distance.