基于k均值聚类和自适应模板匹配的眼底出血点检测方法

Hemorrhages Detection in Fundus Image Based on k-Means Clustering and Adaptive Template Matching

眼底出血点是糖尿病视网膜病变的早期症状,准确检测眼底图像中的出血点,对于构建糖尿病视网膜病变的自动筛查系统具有重要意义,本研究提出了一种基于k均值聚类和自适应模板匹配的出血点检测方法。首先利用HSV空间亮度校正以及对比度受限自适应直方图均衡化方法对眼底图像进行预处理,然后使用k均值聚类分割出候选目标,最后利用自适应归一化互相关模板匹配与支持向量机(SVM)分类器对候选目标进行筛选,从而得到真正的出血区。采用DIARETDB数据库的219幅眼底图像进行实验,本方法在图像水平的灵敏度为100%,特异性为80%,准确率为92.4%,在病灶水平的灵敏度为89%,阳性预测值为87.3%。结果表明本方法能够实现眼底图像中出血点的自动检测。

Hemorrhages are early symptoms of diabetic retinopathy （DR）, the accurate detection of hemorrhages in fundus images is an important contribution for building automatic screening system of DR, a novel algorithm based on k-means clustering and adaptive template matching was proposed in this work. Firstly, HSV brightness correction and contrast limited adaptive histogram equalization were applied to fundus images. Then, the candidate hemorrhages were extracted by using k-means clustering. At last, adaptive template matching with normalized cross-correlation and SVM classifier were used to screen the candidates, and the hemorrhages were detected. The approach was evaluated on 219 fundus images from the databases of DIARETDB. Using an image criterion, we achieved 100% sensitivity, 80% specificity and 92.4% accuracy. With a lesion criterion, we reached a sensitivity of 89% and a positive predictive value of 87.3%. The results show that hemorrhages in fundus images can be detected automatically using this method.