基于局部区域主动轮廓与抠图的图像分割算法  

Image Segmentation Algorithm Based on Local Region Active Contour and Matting

作　　者：曹淼科 方发明 徐盈盈[1] 沈超敏[1] 

机构地区：[1]华东师范大学信息科学技术学院,上海200241

出　　处：《计算机工程》2015年第8期190-195,共6页Computer Engineering

基　　金：国家"973"计划基金资助项目(2011CB707104);国家自然科学基金资助面上项目(61273298)

摘　　要：在复杂边界下分割目标时,经典主动轮廓模型无法对边界做出精细分割,而基于全局信息的分割方法无法对异构图像做出很好分割。为此,基于Chan-Vese模型将抠图方法引入到主动轮廓图像分割算法中,采用局部区域分割的方式,将Chan-Vese模型作为抠图算法的引导项,构建能量泛函。通过变分法求解能量泛函,以迭代的方式求出近似最优解。对比实验结果表明,提出的算法能够精确地分割出目标图像,并且相对于Chan-Vese模型,具有较好的鲁棒性和对初始轮廓的不敏感性。Aiming at overcoming the defaults that classical active contour models cannot segment objects accurately with details,this paper proposes a hybrid approach for natural image segmentation using region based active contours and matting.Since localized contour methods can segment objects with heterogeneous,which is a difficult task for a global method,it combines matting and Chan-Vese models,uses the way of local area segmentation.The closed form matting is used as a trimap in the localized Chan-Vese model and the energy functional is built.It implements the model by variational methods,and obtains the optimal solution by iteration.The comparison results illustrate that the model can extract objects accurately.Compared with Chan-Vese model,it has salient features of robustness and insensitive to initial curve.

关 键 词：主动轮廓 区域信息 抠图 能量泛函 变分法 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]