融合重建滤波器和暗特征的渗出物检测算法  

Retinal exudates detection algorithm by fusing reconstruction filters and dark features

作　　者：赖小波[1] 刘华山[2] 

机构地区：[1]浙江中医药大学信息技术学院,浙江杭州310053 [2]东华大学信息科学与技术学院,上海201620

出　　处：《光电子．激光》2013年第6期1238-1244,共7页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(61203337);浙江省自然科学基金(LQ12F01004);浙江中医药大学校级科研基金(2011ZZ10)资助项目

摘　　要：针对多数视网膜渗出物提取算法检测精度不高的问题,提出了一种融合重建滤波器和暗特征的视网膜渗出物检测算法。首先,对视网膜灰度图像均值滤波后应用"球形"重建滤波器移除视网膜血管,使用列滤波器标识出渗出物区域并阈值分割得到含有视盘和边界的渗出物区域图像。其次,在均值滤波后的灰度图像中检测视盘和图像边界,并根据视盘构建掩膜,接着从含有视盘和边界的渗出物区域图像中移除视盘和边界,然后应用"圆盘形"重建滤波器扩充渗出物尺寸。最后,对视网膜灰度图像自适应直方图均衡化,阈值分割后反转灰度值得到暗特征,从扩充渗出物尺寸后的图像中移除暗特征即得到最终渗出物图像。实验结果表明,本文算法能有效检测视网膜眼底图像中的渗出物,较现有算法具有更高的检测精度。Aiming at the problem that the detection accuracy is not high for most retinal exudates detection algorithms, a retinal exudates detection algorithm fusing the reconstruction filters and dark features is proposed. Firstly, the mean filter is applied to the retinal grayscale image before the ‘ball＇ reconstruction filter is conducted to remove the blood vessels, the column filter is used to mark the exudates reNon,and the exudates region image with the optic disc and borders is obtained after thresholding segmentation. Secondly, the optic disc and image borders are detected in grayseale image with the mean filter applied, and the mask is created based on the optic disk, then the optic disc and borders are removed from the exudates region image containing the optic disc and borders,after which the‘disk＇ reconstruction filter is implemented to expand the exudates sizes. Finally, the retinal grayscale image is applied with adaptive histogram equalization, dark features are extracted with intensity reversion after binarizing the image by thresholding,and the final image of exudates is obtained after removing the dark features from the image with expanded exudates’ sizes. The experiment results indicate that the proposed algorithm can effectively detect the exudates of retinal fundus image,and it also has high detection accuracy.

关 键 词：渗出物检测 重建滤波器 暗特征 视网膜眼底图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]