Abstract: A variable speed image segmentation algorithm based on enhanced Geometric Active Contour（GAC） model is proposed in order to solve the problem that traditional GAC model usually leaks boundary. The algorithm combines image edge gradient information and coordinate information of corners. In the algorithm, the constant speed of evolution curve on the corners and weak boundary is changed to avoid active contour curve continue evolution into the target boundary, causing boundary leakage and corner loss, influencing the accuracy of target contour extraction. The experimental results show that the proposed algorithm can make the evolution curve stop at the edge of target more accurately and reach an obvious effect in terms of boundary leaking, in comparison with traditional GAC model.

Key words: geometric active contour model, level set, image segmentation, corner extraction, boundary leaking

摘要： 针对传统几何活动轮廓（GAC）模型易出现边界泄露的缺陷，提出一个基于改进GAC模型的图像变速分割算法。该算法结合了图像边缘梯度信息和边缘角点坐标信息，通过改变演化曲线在角点及弱边界处的常量速度，避免活动轮廓曲线继续演化进入目标边界内，造成边界泄露和角点丢失现象，影响目标轮廓提取的准确性。实验结果表明：该算法可使演化曲线更加准确地停在目标边缘，并且在一定程度上减少了边界泄露问题。

关键词: 几何活动轮廓模型, 水平集, 图像分割, 角点提取, 边界泄漏