基于视觉显著性的层次形状分解方法  

A Visual Salience Based Hierarchical Shape Decomposition Algorithm

作　　者：马燕新 鲁敏[1] 滕书华[1] 张军[1] 

机构地区：[1]国防科学技术大学电子科学与工程学院,长沙410073

出　　处：《计算机辅助设计与图形学学报》2014年第6期914-922,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(60972114);中国博士后科学基金(2012M512168)

摘　　要：形状分解是形状理解与分析的重要基础.在充分理解人类视觉特性的基础上,联合形状骨架与轮廓特征构建了一种视觉显著性度量即视觉显著度,并提出一种基于视觉显著度的层次形状分解方法.首先采用基于距离变换的骨架生成方法获取目标骨架,之后利用骨架分叉点生成所有候选分割线,最后通过分叉点对应的内切圆半径以及轮廓段的视觉显著度对分割线进行优选获得最优解.实验结果表明,该方法对噪声、形变具有较好的鲁棒性,分解结果符合人类视觉习惯;此外,通过调整显著性阈值可获得不同尺度下的细节分解结果,具有更好的灵活性.Shape decomposition is an important foundation for shape understanding and analysis.Based on the understanding of the characteristics of human vision,we proposed a measurement of human vision named visual salience which combines the strength of both skeleton and boundary features.And then a hierarchical shape decomposition method is proposed.First,the skeleton of a shape is produced using a robust distance transform based algorithm.Then,all possible partition lines are retrieved by analyzing the junction points on the skeleton.Next,the visual salience and the radius of the inscribed circle of the junction point are used to discard redundant partition lines.The remaining partition lines are selected to decompose the shape.Experiments show that our method satisfies subjective visual perception on shape decomposition and is robust to large shape noise.Furthermore,our method is flexible and produces controllable decomposition results.

关 键 词：形状分解 骨架提取 视觉显著度 形状层次分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]