【摘要】 眼底图像视盘定位是视盘分割的重要前提.针对视盘定位结果易受图像对比度的影响的问题,提出一种基于置信度计算的快速视盘定位方法.首先采用基于形态学变换的方法增强眼底图像中视盘、血管区域与图像背景的对比度,并根据图像增强结果中像素点的亮度特征初始定位视盘区域;然后运用局部滑动窗口扫描的方法,根据窗口内像素点亮度特征和其周围血管分布的特性计算候选区域的置信度,定位视盘区域.在不同的眼底图像公共数据上进行实验的结果表明,对于1 341幅眼底图像,该方法能准确地定位其中1 325幅图像的视盘区域,视盘定位准确率为98.8%,平均每幅图像耗时0.25 s,优于现有的视盘定位方法,适用于眼底疾病的计算机辅助诊断.更多还原

【Abstract】 The optic disk localization is an important step before disk segmentation of fundus images. Since the optic disk localization can be easily influenced by the image contrast, we propose a fast optic disk localization method based on confidence score computation. Firstly, we perform morphological transformations to enhance the contrast between the optic disk and vessel regions from the background. According to the intensity features of pixels in the enhanced image, we localize the disk region roughly. Then, we apply the local sliding window scanning method to roughly localize the disk region and compute the confidence score of each candidate region based on the pixel intensity features in the window and the neighborhood vessel properties to generate the optic disk localization result. We conducted experiments on several publicly available datasets. The results show that the proposed method can successfully localize 1 325 optic regions among 1 341 images, and achieves 98.8% accuracy of the optic disk localization with each image consuming 0.25 s on average. The proposed method outperforms other existing optic disk localization methods and can be used in computer-aided eye disease diagnosis.更多还原