视盘自动分割算法

Optic disc automatic segmentation algorithm

针对多数视盘提取算法因不均匀光照导致检测精度不高的问题,提出了1种基于阴影校正的视盘自动分割算法。首先,利用Otsu方法分割视网膜红色分量眼底图像,提取和填充最大连通区域,均值滤波后得到视网膜掩膜图像,并与视网膜红色分量眼底图像卷积;其次,运用特定形式的均值滤波器平滑视网膜图像边界点,创建视网膜阴影图像;最后,将卷积图像减去阴影图像,再次利用Otsu方法进行分割,提取最大连通域并填充,应用"圆盘形"结构元素进行形态学"开"运算提取最终的视盘图像。将所提出的算法在视网膜眼底图像有1 200张的MESSIDOR数据库和40张的本地数据库分别进行验证。实验结果显示,所提出的算法平均重叠率分别为0.895和0.91,高于目前相关视盘分割算法,表明即使在非均匀光照条件下也能获得满意的分割结果,基本能够满足临床应用对稳定性和精度的要求。

Aiming at the problem that the detection accuracy is not high for most optic form illumination, an optic disc automatic segmentation algorithm based on shading correction is proposed. Firstly, the Otsumethod is used to segment the retinal red component image, and maximum connected region mask ^mage s obtained after mean filtering, and s convolved with the retinal red component specific form s applied to smooth the boundary points m retiml ^mage, after which the retirnl mask image s created. Finally, by subtracting the shade-corrected image from the convolution image, then maximum connected region s extracted and filed after the Otsu segmentation, and the 4open？ operation with 4disk-shaped？ structuml element s image. The experimental results rnidicate that the average overlap rate of the algorithm on MESSIDOR database nicluding 1200 images s 0. 895 , and on the local database rnicluding 40 images s 0. 91 , which are higher than other optic disc segmentation algo-rithms. tt shows that satisfactory segmentation results also can be obtained even under nonuniform illumination conditions, which satisfies the clinical application requirements for strong stabilization and higher precision.