基于显著性引导的保凸性水平集分割算法

A Level Set Segmentation Algorithm Based on Saliency Guidance and Convexity Preservation

结合彩色眼底图像中视神经盘的解剖学特征,引入零水平集曲线的曲率变化来优化活动轮廓模型,提出了一种全自动分割视神经盘的水平集算法CRLSF。该算法利用显著性算法快速获取初始轮廓,且该初始轮廓与视神经盘轮廓非常接近,大幅度提高分割效率。采用保凸机制能够使得曲线在演化过程中不受到视神经盘周围毛细血管的影响以保持分割曲线的凸性。最后,在迭代中引入目标函数来评估算法的收敛性。实验表明,提出的CRLSF算法优于已有的算法。

Combined with the anatomical features of the optic nerve disc in color fundus images,the curvature change of the zero-level set curve is introduced to optimize the active contour model,and a horizontal set algorithm for fully automatic segmentation of the optic nerve disc is proposed.The algorithm uses the significance algorithm to quickly obtain the initial contour,and the initial contour is very close to the optic disc contour,which greatly improves the segmentation efficiency.In addition,the convex preservation mechanism can make the curve not be affected by the capillaries around the optic disc during the evolution process to maintain the convexity of the segmentation curve.Finally,the objective function is introduced in the iteration to evaluate the convergence of the algorithm.Experiments show that the proposed CRLSF algorithm is superior to the existing algorithm.