提出了一种利用多尺度斑点检测的快速视盘定位方法。首先通过斑点检测确定12个视盘的候选区域, 然后综合利用候选点邻域内的外观和血管特性决策视盘真实位置。对五个公开的数据集1540幅图像进行了测试, 定位准确率达到97.9%。算法可以在缩小后的图像上进行检测, 因此检测效率高, 在五个数据集中平均检测时间约0.2s, 远快于目前绝大部分检测算法。实验表明, 对正常和轻度病变图像视盘检测效果很好, 但是对严重病变的图像以及部分低质量的图像中的视盘检测, 准确率仍有待提高。

An efficient method to detect the optic disc(OD) is proposed based on multi-scale blob detection algorithm. By blob detection, twelve candidate optic disc regions are determined in advance. Then by comprehensive utilize the appearance and vascular characteristics of the OD candidate points, the true position of the OD can be distinguished. In the experiment, 1540 images collected from five public datasets are tested, and the overall OD localization accuracy is 97.9%. Because the algorithm is implemented in a smaller resized image, it has a high efficiency. Average computation time for the data sets are about 0.2s/image, which is much faster than most of present detection algorithms. Experimental results show that the method has excellent detection performance for normal images or images with mild lesions, but in some serious diseased or low-quality images, the accuracy is still needs to be improved.