基于2D人体图像特征学习的女西装合体性判别  

Discriminating the fit of women’s suits based on 2D human body image feature learning

作　　者：姚彤[1,2,3] 闵悦宁 王军 孙见梅[3] 潘力 YAO Tong;MIN Yuening;WANG Jun;SUN Jianmei;PAN Li(School of Fashion,Dalian Polytechnic University,Dalian 116034,China;National Demonstration Center for Experimental Fashion Design and Engineering Education,Dalian Polytechnic University,Dalian 116034,China;School of Textile and Material Engineering,Dalian Polytechnic University,Dalian 116034,China)

机构地区：[1]大连工业大学服装学院,大连116034 [2]大连工业大学服装设计与工程国家级实验教学示范中心,大连116034 [3]大连工业大学纺织与材料工程学院,大连116034

出　　处：《丝绸》2024年第5期78-86,共9页Journal of Silk

基　　金：辽宁省教育厅项目(LJKMR20220912,LJKZZ20220068,JYTMS20230401)。

摘　　要：为解决服装远程定制和网络购衣过程中服装合体性无法预先判断的问题,文章提出利用2D人体图像判断女西装合体性。首先采集了462名18~25周岁青年女性的2D图像及3D尺寸,建立人体数据集。并制作10件虚拟女西装,建立服装数据集。其次提取用于女西装合体性判别的服装特征、2D人体图像关键点距离特征和整体特征。基于感性工学原理,采用专家评价法获取女西装合体性评价标签。最后利用贝叶斯分类器建立女西装合体性判别模型。研究结果显示,将文章提出的特征应用于基于贝叶斯算法的女西装合体性判别模型判别准确率可达到84.8%,说明提出的服装合体性评价指标有效,可以用于基于2D图像的服装合体性预测。Garment fit evaluation is one of the main bottlenecks in current fashion design and manufacturing.In the process of remote customization and online shopping,the size of the clothes that does not fit has become the primary reason for users to return or exchange them,which has seriously affected the intelligent upgrade of the clothing industry.Therefore,how to judge the fit degree of a garment without actually trying it on is one of the main problems that needs to be solved urgently in the clothing industry.In previous studies,garment fit was usually evaluated by some characteristic indexes,such as ease allowance,clothing wrinkle,the spatial relationship between clothing and the human body,and clothing pressure.However,most of them have some problems such as the need to try on clothes and the complicated evaluation process,which are not suitable for remote customization of clothing and online shopping.With the development of computer vision and machine learning technology,a small number of scholars use 2D images to judge garment fit based on clothing wrinkle,but only for specific garments,and the generalization ability is weak.To solve the problem that garment fit cannot be discriminated in advance in the process of remote customization and online shopping,this study took women’s suits as an example and put forward a method to discriminate the fitness of women’s suits based on 2D human images.Firstly,2D images and 3D human body size of 462 young women aged 18-25 in northeast China were collected,and a human body data set was established.Twenty-two samples were selected from this data set,and virtual models were established for virtual try-on by using CLO3D software.According to the national clothing size standard of China,10 virtual women’s suits with the same style and different sizes were made to establish the garment data set.Based on the principle of Kansei engineering,the fitting evaluation labels of women’s suits,namely tight,fit and loose types were given by expert evaluation method combined with vir

关 键 词：2D人体图像特征 计算机视觉 机器学习 女西装 合体性 贝叶斯模型 

分 类 号：TS941.17[轻工技术与工程—服装设计与工程]