形状填充率的物体识别算法  

Object Recognition Using the Shape Filling Rate

作　　者：安国成[1] 张凤军[1] 戴国忠[1] 

机构地区：[1]中国科学院软件研究所人机交互技术与智能信息处理实验室,北京100190

出　　处：《计算机辅助设计与图形学学报》2010年第6期1016-1022,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(60673188;60672094;U0735004);国家"八六三"高技术研究发展计划(2008AA01Z303);国家"九七三"重点基础研究发展计划项目(2009CB320804)

摘　　要：针对经典形状上下文算法对物体关节相对位置变化敏感的缺点,提出一种基于剪影局部形状填充率的物体识别算法.该算法以物体不同的轮廓控制点为圆心,计算不同半径下物体剪影像素所占总像素的比例,即为控制点的局部形状填充率;将不同控制点、不同半径长度所计算的形状填充率数值构成一个特征矩阵,该矩阵反映了物体整个剪影的统计特性.通过不同数据库的实验结果表明,文中算法对物体的细节有很强的描述能力,对物体关节的相对位置不敏感,并且受剪影轮廓控制点数量影响小.The shape context algorithm is sensitive to the variance of the articulation relative position.A shape recognition method is proposed in this paper by using the information of the local shape filling rate of object silhouettes.The ratio of the shape silhouette pixels over the total pixels is defined as the local shape filling rate and it takes the landmark point as the center of circle having different radius.The shape filling rates of all the landmark points with different radius constitute a characteristic matrix that reflects the object entire statistical property.The new approach is tested on a variety of shape databases.The novel method has strong power to describe the object details.It is insensitive to articulation and the quantity of landmark points.

关 键 词：形状上下文 内部距离 直方图 形状填充率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]