中文摘要：

基于水平集方法和结构张量,提出几何活动轮廓模型应用于图像分割,解决水平集方法轮廓初始化和弱边缘处易于边缘泄露问题。该方法利用张量图像的散度算子构造新的外力,引导水平集函数的自适应运动,使得其可以初始为常值函数,消失其演化对初始轮廓的需要;在偏微分方程中引入张量迹信息,减少噪声对其演化的影响,避免轮廓在弱边缘处泄露。实验结果表明,该方法对噪声图像鲁棒,能提取深度凹陷目标轮廓和红外图像中的弱目标。

英文摘要：

This paper proposed a geometric active contour model based on the level method and structure tensor for image segmentation. The goal was to deal with the problems of contours initialization and the leak of weak boundaries. With the proposed model,it presented a new force via the divergence operator of the tensor-image,which guided the level set function moving up and down adaptively. So the level set function could be initialized to a constant function,which completely eliminated the need of initial contours. By incorporating the information of tensor trace into the common partial differential equation,the regularization on zero level curves could diminish the influence of image noise on level set evolution while ensuring the active contours not to pass through weak object boundaries. Experimental results show that the proposed method is robust to noising image,and can extract the concave boundaries and the weak objects in infrared image.