基于边缘交点聚类法的眼底图视盘轮廓提取

Optic Disc Contour Extraction from Retinal Images Based on Edge Intersect Points Clustering

由于眼底血管闭塞、噪声和弱边界等原因,主动轮廓模型不能够精确地收缩至视盘的轮廓边缘,由此提出基于边缘交点聚类的轮廓提取方法。由初始轮廓确定一个矩形区域,选取该矩形区域的主对角线、副对角线、垂直中心线和水平中心线,求其与视盘边缘的交点,将所有候选交点进行K-Means聚类分析,取得位于边缘附近的点作为主动轮廓模型的初始轮廓点,进行能量最小化计算,直至轮廓停止收缩。实验证明,经边缘交点聚类初始化后再运用主动轮廓进行视盘边缘提取,轮廓最终会较为精确地收缩于视盘边缘。

The active contour model can not converge to the optic disc boundary precisely because of the retinal vascu- lar occlusion,noise and weak boundaries and so on,so a contour extraction method is proposed based on clustering of the edge intersect points. First, a rectangle was determined by the initial contour,and then the principle diagonal, vice diagonal, vertical center line and horizontal center line were selected to compute their intersect points with the optic disc boundary,which were then clustered by K-Means algorithm, with the points around the optic disc boundary as the initial contour of the active contour model obtained. Second, the energy was minimized until the contour stopped converging. The experiment shows that after initialization by the edge intersect points, the optic disc boundary ex- traction by the active contour model is more precise and efficient.