活动轮廓模型基于结构张量的边缘停止函数

Edge stopping function based on structure tensor for active contours

所有边缘活动轮廓模型都依赖边缘停止函数,该函数通常是高斯平滑图像的梯度模的单减函数。梯度能够刻画图像的局部边缘特征,但忽略了边缘的"分叉点"、"角点"等重要信息,这导致了边缘定位不准甚至产生错误的分割。基于图像结构张量的一个局部相干性度量,提出一个新边缘停止函数。实验结果表明,基于这个边缘停止函数,活动轮廓模型能够精确定位目标边缘,同时大大减少了迭代次数并具有较强抗噪性。

All of edge-based active contour models rely on edge stopping function.It is typically a decreasing function of the gradient magnitude of Gaussian smoothed image.In identifying object edges,however,image gradient does not take into account some important features such as junctions and corners;this results in the inaccurate location of edges or even false segmentations.Based on a measure of local coherence of image structure tensor,this paper proposes a new edge stopping function.Experimental results show that an active contour model using the new edge stopping function can significantly perform better in the location of edges,while it is much faster and more robustness to noise than the original model.