一种基于多特征的距离正则化水平集快速分割方法

摘要

现有的图像分割模型存在对初始化信息敏感,分割速率慢,图像弱边界区的泄露等现象.提出了一种混合快速分割方法.该方法利用偏压场近似估计图像的局部统计信息,并结合全局信息相容性及改进的距离正则化方法建立模型,最后将模型嵌入水平集框架中,与此同时,引入双重终止准则以提高分割的速度.最后利用合成图像和真实图像进行分割实验,并与CV (Chan-Vese)模型、非线性自适应水平集方法以及局部尺度拟合模型对比,表明本方法不仅对初始化信息敏感度降低,而且分割速度提高3～5倍.%The existing image segmentation models have problems of being sensitive to initialization information,slower segmentation and leaked weak image boundary regions.This paper presents a hybrid fast segmentation model which utilizes the local statistics of bias field approximated images,the global information of compatibility and the distance regularization method.Then the model is embedded into level set framework.In addition,a dual termination standard is constructed to improve the speed of segmentation.Experiments on synthetic and real images are conducted to verify the efficiency of our model.Moreover,comparisons with the well-known CV model,nonlinear adaptive level set model and region scalable fitting model demonstrate that the proposed model reduces the sensitivity to the initialization and improves the segmentation speed by 3 ～ 5 times.