基于Morlet小波变换的视网膜血管分割

Using Morlet Wavelet for Retinal Vessel Segmentation

眼底视网膜血管网络是诊断糖尿病视网膜病、青光眼等眼科疾病的重要手段.根据视网膜血管的树状网络结构和灰度分布特征,提出一种基于Morlet小波和高斯匹配滤波的分割方法.首先通过分析二维Morlet小波变换对血管的系数响应来构造血管特征函数图;随后采用多尺度的离散高斯核对血管骨架进行匹配滤波,提高微小血管与背景区域的对比度;最后结合区域连通性分析和滞后阈值技术滤除背景噪声,提取出更加精确的血管树细节.在DRIVE和STARE数据库上的实验结果表明,作为非监督类分割方法,该方法能有效地提取眼底图像的视网膜血管网络,粘连现象少,而且对图像中噪声的鲁棒性较其他方法明显提高,具有较好的临床应用参考价值.

Extracting retinal vessels from the retinal image is important to the diagnosis of diabetic retinopathy and glaucoma in clinic. In terms of special gray distribution and regional configuration of retinal blood vessel network, this paper proposed a new blood vessel tree segmentation method based on Morlet wavelet and Gaussian matched filter. To construct the feature vector image, the Morlet wavelet cofficient responses are computed for the retinal vessel. Length filtering and hysteresis thresholds are subsequently implemented for the vessel tree extrac- tion. The experimental results show that the proposed unsupervised method can effectively extract the blood ves- sels network on fundus image. Compared with other methods, our method has stronger robustness with less vessel stickiness, which is more practically valuable for the clinic diagnosis.