关键词: 图像分割, 活动轮廓模型, 边缘停止函数, 结构张量, 局部相干度量

Abstract: 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.

Key words: image segmentation, active contour model, edge stopping function, structure tensor, local coherence