基于背景估计和集成分类的眼底硬性渗出检测

Detection of hard exudates in retinal images based on background estimation and ensemble classification

眼底图像中硬性渗出的检测对于糖尿病性视网膜病变(简称"糖网")的早期诊断具有重要意义。为了实现眼底图像中硬性渗出物的自动检测,本文提出了一种结合背景估计和集成分类的眼底图像硬性渗出物自动检测方法。首先,对图像进行亮度校正、去噪、对比度增强等预处理操作,然后结合形态学方法进行背景估计和图像重建并移除视盘区域,得到渗出物候选区域。最后利用集成分类方法对候选区域进行分类,提取最终的硬性渗出区域。实验结果表明,本方法能够有效准确地检测到眼底图像中的硬性渗出物,对于糖网自动诊断技术的研究具有积极意义。

The detection of hard exudates in retinal images is an important task to the early diagnosis of diabetic retinopathy（ DR）. This paper presents a novel method to automatically detect hard exudates in retinal images,which is based on background estimation and ensemble classification. In the first stage,retinal images are preprocessed for shade correction,denoising and contrast enhancement. Then the candidate hard exudates are extracted by background estimation,morphological reconstruction and elimination of the optic disc. At last,an ensemble classifier is trained for the classification of candidate regions. As the experimental results suggest,the proposed method can detect hard exudates in retinal images accurately,which is a contribution to the research of the automatic diagnosis technique for DR.