基于置信度计算的快速眼底图像视盘定位

发布时间：2018-08-26 15:33

【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.
[Abstract]:The location of visual disc in fundus image is an important premise of disc segmentation. Aiming at the problem that the result of disk localization is easily influenced by image contrast, this paper presents a fast method based on confidence degree calculation. Firstly, the contrast of visual disc, vascular region and image background in the fundus image is enhanced based on morphological transformation, and the initial location of the disc region is based on the luminance characteristics of pixels in the image enhancement result. Then using the method of local sliding window scanning, the confidence degree of candidate region is calculated according to the luminance characteristics of pixels and the distribution of blood vessels around the window, and the optical disk area is located. The experimental results on common data of different fundus images show that the method can accurately locate the optical disk region of 1,325 of the 1,341 fundus images. The accuracy of disc localization is 98.8, and the average time of each image is 0.25 s, which is superior to the existing method of optical disk localization and is suitable for computer-aided diagnosis of ocular fundus diseases.
【作者单位】：中南大学信息科学与工程学院;有色金属工业信息化与自动化工程技术研究中心;中南大学湘雅二医院眼科;
【基金】：国家自然科学基金(61573380,61562029,61672542)
【分类号】：TP391.41

【相似文献】