基于视觉注意机制的眼底图像视盘快速定位与分割

Fast Location and Segemention of Optic Disk in the Fundus Image Based on Visual Attention

视盘定位与分割对利用眼底图像进行眼科计算机辅助诊断十分重要。视觉注意机制模拟人类视觉系统从而能在复杂场景中快速定位目标。利用视盘在眼底图像中视觉显著度高这一特性,提出基于视觉注意模型实现视盘的快速定位并分割的算法。算法首先对不同眼底图像进行归一化,接着采用高斯金字塔提取图像不同尺度的颜色、亮度和方向特征图(feature map),进一步整合得到显著性图(saliency map),在显著性图中提取FOA(focus of attention)从而定位视盘。接着在视盘定位区域采用统计排序滤波器(rank-filter)抹除血管,在极坐标中采用亚像素提取图像边缘,实现视盘分割。采用国际Messidor数据库来验证算法的性能,定位精度为95%,运行时长为0.213 s,与其他算法相比,算法具有准确性高和实用性强等特点,具有良好的应用潜力。

The location and segmentation of optic disc is very key to the computer-aided diagnosis of ophthalmology by the fundus image. The mechanism of visual attention which simulates human visual system can quickly locate the target in a complex scene. Using high saliency of the disc in the fundus image,we propose a method which based on visual attention model to quickly locate and segmente the disc. Firstly,different fundus image is normalized. Then Gaussian pyramid is used to extraction color feature maps at different scales,as long as intensity and orientation feature maps,further integrate to generate the saliency map. The disc finally is located by extract focus of attention in the saliency map. After that we erase the vessels using rank filter in the region around the disc,extract the sub-pixel edge in polar coordinates,and finally carry out the segmentation of optic disc. In this paper,we verify the performance of the algorithm by international Messidor database,the accuracy of location is 95%,the mean of run time is 0. 213 seconds. Compared with other algorithms,the propose algorithm has high accuracy and practical,and has a good potential in the future.