摘要
尺度可控的局部区域(RSF)活动轮廓模型可用于灰度不均匀图像的分割,但存在初始化敏感和易陷入局部极小值的缺点,从而限制了其实际应用,因此提出了一种结合模糊聚类区域信息的变分水平集活动轮廓模型.该模型采用了模糊均值聚类(FCM)算法对图像进行预处理,将预处理结果二值化后作为下一步水平集演化的初始轮廓,解决了初始化敏感问题;设计了一个灰度域上的核函数,将其与RSF模型的空域核的一个线性组合作为局部能量项,弥补了采样权值仅与空间距离有关的缺陷,提高了分割精度;将聚类分析得到的模糊隶属度作为图像的全局信息,结合改进的CV模型,作为全局拟合力,增加了模型的自适应性,并加快了模型的收敛速度.实验结果表明,该模型能够自动初始化,抗噪性能强,对灰度不均匀图像有很好的分割效果.
Abstract：The Region-Scalable Fitting （RSF） model can be used to segment images with intensity inhomogeneity, but it is so sensitive to the location of initial curve and is easy to fall into local minimums that limits its practical application. Therefore, a region-based ac- tive contour model combining the fuzzy means clustering （FCM） in a variational level set formulation is proposed in this paper. In the proposed method, it uses the fuzzy means clustering algorithm for image preprocessing, and the results of binarization of the pre- processing serves as the initial contour of the Next level set evolution, which solves the sensitivity to the location of initial curve; The local energy item is defined as a linear combination of the RSF model and our model by taking domain and range kernel functions into account, which can make up for the defects that sampling weights are only related to spatial distance and improve the accuracy of seg- mentation. Also, it establishes a global fitting force by introducing the fuzzy membership of the Cluster analysis that serves as the global information of image and combine with the improved CV active contour model, which can increase the self-adaptability of the model and can also accelerate the speed of convergence of the proposed model. Experimental results show that the proposed model al- lows for automatic initialization and is less sensitive to noise, while it has been applied to images with intensity inhomogeneity with desirable results.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第3期663-666,共4页
Journal of Chinese Computer Systems
基金
河北省卫生厅科研基金项目(20120395)资助
关键词
图像分割
活动轮廓模型
模糊均值聚类
水平集函数
灰度不均
初始轮廓线
image segmentation
active contour model
fuzzy means clustering
level set function
intensity inhomogeneity
initial contour
