用于图像分割的鲁棒的区域活动轮廓模型

Robust Region-based Active Contours Model for Image Segmentation

针对非同质或者弱边界图像分割时出现的问题,提出一种改进的活动轮廓模型。首先,由图像的区域统计信息定义了一个新的能量泛函。区域统计信息由局部信息和全局信息采用新的加权组合而成。其次,采用水平集方法最小化该能量泛函,得到水平集演化方程并不断更新。最后,采用高斯滤波方法规则化水平集方程。此外,该模型可以退化成一种无需初始化和规则化的简单的全局活动轮廓模型。合成图像和真实图像的实验结果表明:该模型能有效地分割非同质或弱边缘图像,对噪声并初始轮廓曲线具有较好的鲁棒性,并且计算效率高。

A novel region based active contours model was proposed to deal with the images with intensity inhomogeneities and weak boundaries.For the proposed model,a new energy function was defined,which consists of a local fitting term and an auxiliary global fitting term.Then,the energy functional was incorporated into a variational level set formulation.Furthermore,we regularized the level set function by using Gaussian filtering to keep it smooth and eliminated the re-initialization.In addition,the proposed model can degrade to a new global CV model.Experiments results show that the proposed model can not only segment images with intensity inhomogeneities and weak boundaries,but also robust to the noise and initial contours.Also,it has high computational efficiency.