梯度向量场通量能量驱动的主动轮廓模型

Active Contours Driven by Image Gradient Flux Energy

提出了一种基于梯度向量场通量能量的水平集图像分割算法。通过加入约束符号距离函数的能量项,并极小化该能量函数得到的变分表达式主要具有4条优于传统主动轮廓模型的优点。一是可以克服分割弱边界目标的困难;二是水平集函数不但可以灵活初始化,而且可避免在演化过程中重新初始化为符号距离函数;三是水平集函数数值化可采用简单的有限差分方法,计算效率得到了极大的提高;四是仅用一个初始轮廓就可以自动检测带孔目标的内轮廓。对合成和真实图像的分割结果表明:对弱边界目标和灰度分布不均目标的分割效果分别优于测地线模型(GAC)和C-V主动轮廓模型。

A new image segmentation algorithm using level set method was presented, which is based on the gradient vector field flux fitting energy. Combing with the penalizing energy term of signed distance function and minimizing the energy functional, we obtain the variational formulation which has four main advantages over the traditional active contour models. First, the difficulty of segmentation with weak edge can be addressed. Second, the level set function allows for flexible initialization and needs no re-initialization during evolution. Third, the level set function can be easily implemented by simple finite difference scheme and is computationally more efficient. Fourth, the interior contour of object can be automatically detected with only one initial contour. The proposed algorithm has been applied to both synthetic and real images with promising results and the results of segmenting weak edge objects and images with intensity inhomogeneity are better than using either geodesic active contour （CAC） model or C-V active contour model, respectively.