基于体表角度的女子体型分类与识别  

Classification and Automatic Identification of Females' Body Shape Based on Body-Surface Angles

作　　者：孙洁[1] 倪世明[1] 叶玲[1] 邹奉元[1,2] 

机构地区：[1]浙江理工大学服装学院,杭州310018 [2]浙江理工大学浙江省服装工程技术研究中心,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2013年第2期184-188,共5页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：国家茧丝绸发展专项项目(财企[2011]181号);浙江省实验教学示范中心建设项目(SB1105001-E)

摘　　要：从人体形态差异角度细分人体体型,通过三维人体扫描获得500名年龄在18～25岁之间的女青年的正面和侧面二维图像,运用Matlab软件获得11个躯干体表特征点的坐标并计算得到能够反映体型差异的体表角度;通过偏相关性分析得出表征人体形态特征的4个体表角度,即肩斜角、胸突角、体侧角、臀突角;运用K-means聚类将人体体型分为4类。在此基础上,构建基于神经网络集成的体型识别模型,训练集识别精度达到95%。该方法可有效区分人体形态差异,适应服装量身定制生产的需求。To classify the body shape of human from the perspective of morphological differences of human body, front and side two-dimensional images of 500 young girls between 18 and 25 are obtained through three-dimensional body scan. This paper obtains coordinates of 11 body-surface feature points by using Matlab software and calculates body-surface angles reflecting body difference; obtains 4 body-surface angles reflecting morphological characteristics of human body through partial correlation analysis, i. e. shoulder oblique angle, breast coign, side angle and buttocks coign; and classifies human body shape into four types by using K-means clustering. A body shape identification model based on neural network ensemble if established on this basis and the training set identification accuracy reaches 95%. This method can effectively distinguish morphological differences of human body and adapt to the requirement of customized production of clothes.

关 键 词：体表角度 K-MEANS聚类 神经网络集成 人体体型识别 

分 类 号：TS941.17[轻工技术与工程—服装设计与工程]