基于多模板匹配的局部自适应区域生长法在视网膜内出血自动检测中的应用

为自动检测出眼底图像中的视网膜内出血，从而构建基于眼底图像的糖尿病视网膜病变自动筛查系统，提出了基于多模板匹配的局部自适应区域生长法用以自动检测该病灶。首先，对眼底主要生理结构进行光谱特征分析，从而为不同分割目标选取合适的RGB通道;其次，利用HSV空间的亮度校正以及对比度受限自适应直方图均衡方法对眼底图像进行预处理;在此基础上利用设计好的多个模板对图像进行归一化互相关模板匹配获取该病灶候选区域;然后，从中去除视盘、血管以消除相关假阳，从而得到区域生长所需种子;最后，利用局部自适应区域生长法获取其精确轮廓，从而实现该病灶的准确检测。利用该算法对90幅不同颜色、不同亮度、不同质量、不同分辨率眼底图像进行该病灶的自动检测，实验结果表明：该算法能快速、有效地自动检测出眼底图像中的视网膜内出血，且算法稳定可靠，可满足临床需求。

Algorithm of Locally Adaptive Region Growing Based on Multi-Template Matching Applied to Automated Detection of Hemorrhages

In order to automatically detect hemorrhages in fundus images, and develop an automated diabetic retinopathy screening system, a novel algorithm named locally adaptive region growing based on multi-template matching was established and studied. Firstly, spectral signature of major anatomical structures in fundus was studied, so that the right channel among RGB channels could be selected for different segmentation objects. Secondly, the fundus image was preprocessed by means of HSV brightness correction and contrast limited adaptive histogram equalization(CLAHE). Then, seeds of region growing were founded out by removing optic disc and vessel from the resulting image of normalized cross-correlation(NCC)template matching on the previous preprocessed image with several templates. Finally, locally adaptive region growing segmentation was used to find out the exact contours of hemorrhages, and the automated detection of the lesions was accomplished. The approach was tested on 90 different resolution fundus images with variable color, brightness and quality. Results suggest that the approach could fast and effectively detect hemorrhages in fundus images, and it is stable and robust. As a result, the approach can meet the clinical demands.