基于有向复杂网络模型的形状描述与识别  

Shape Description and Recognition Based on Directed Complex Network

作　　者：汤进[1,2] 郅大鹏[1] 江波[1] 罗斌[1,2] 

机构地区：[1]安徽大学计算机科学与技术学院,合肥230601 [2]安徽省工业图像处理与分析重点实验室,合肥230039

出　　处：《计算机辅助设计与图形学学报》2014年第11期2039-2045,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家科技支撑计划(2012BAH95F01);国家自然科学基金(61202228);安徽省科技厅自然科学基金面上项目(1308085MF97);安徽省高等学校省级自然科学研究重点项目(KJ2013A007)

摘　　要：针对传统复杂网络方法对形状的非刚性变形较为敏感等问题,在形状内部距离的基础上利用有向复杂网络进行形状分析.首先提取形状边界点作为复杂网络的节点,以形状边界点之间的内部距离作为复杂网络中节点之间的边权值构建初始网络;然后对初始复杂网络进行k近邻演化,得到不同演化时刻的有向子网络;最后提取各有向子网络的特征来实现复杂网络的特征描述,进而实现形状的特征表示.实验结果表明,该方法对常见的形状变形具有更强的鲁棒性;与传统的无向网络模型相比,具有更高的检索和分类精度.For shape boundary becomes instable in some non‐rigid transformation and other issues on traditional undirected complex network models ,a new directed complex network model based on inner distances has been proposed to characteristic shape boundaries . Firstly , boundary points and inner distances between these points are represented as nodes and weights of edges of the initial network , respectively .Then ,this initial network evolved based on the k‐nearest neighbor method and its sub‐networks were generated at each evolution stage .Finally ,features of these directed sub‐networks are computed and concatenated to describe the shape boundary . Experimental results on both shape recognition and retrieval show that the proposed method can perform more robust than traditional undirected complex network models .

关 键 词：复杂网络 内部距离 形状描述 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]