服装风格图像的识别分类研究进展  

Research progress in recognition and classification of clothing style images

作　　者：张琦妍 支阿玲[2] 罗戎蕾[3,4] ZHANG Qiyan;ZHI Aling;LUO Ronglei(School of Fashion Design&Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China;School of International Fashion Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China;Silk and Fashion Culture Research Center of Zhejiang Province,Zhejiang Sci-Tech University,Hangzhou 310018,China;Style 3D,Hangzhou 310018,China)

机构地区：[1]浙江理工大学服装学院,浙江杭州310018 [2]浙江理工大学国际时装技术学院,浙江杭州310018 [3]浙江理工大学浙江省丝绸与时尚文化研究中心,浙江杭州310018 [4]浙江凌迪数字科技有限公司,浙江杭州310018

出　　处：《染整技术》2023年第9期1-8,共8页Textile Dyeing and Finishing Journal

基　　金：浙江省一般软科学研究计划项目(2022C35099);浙江省丝绸与文化艺术研究中心培育项目(ZSFCRC20204PY)。

摘　　要：针对服装风格识别与分类,从服装相关数据集、服装图像特征提取、应用场景3个方面综述了该领域的研究进展。服装图像分类数据来源主要包含社交网站、购物网站以及时尚资讯媒体,从季节、时尚风格、服装信息等服装属性归纳数据集特点,围绕图像处理技术对其应用于服装领域最广泛的特征提取方法加以介绍,详细综述了传统特征提取方法、卷积神经网络(CNN)、图卷积神经网络(GCNN)、无监督学习方法在服装领域的相关研究及其模型优化效果、优缺点等。最后分析讨论了服装风格识别分类在时尚风格流行趋势预测、服装搭配推荐方面的发展趋势、关键技术及应用前景。研究认为,未来可以在人体相关特征提取、多角度时尚趋势预测、服装推荐主观评价等方向进一步研究。For clothing style recognition and classification,the research progress in this field is summarized from three aspects:clothing-related data sets,clothing image feature extraction,and application scenarios.The data sources for clothing image classification mainly include social networking sites,shopping sites,and fashion information media.The characteristics of the data set from clothing attributes such as seasons,fashion styles,and clothing information are summarized.The most widely used feature extraction method in the clothing field is introduced around image processing technology,and the relevant research on traditional feature extraction methods,convolutional neural networks(CNN),graph convolutional neural networks(GCNN),and unsupervised learning methods is summarized in detail.Then,the optimi⁃zation effect,advantages and disadvantages of these models are summarized.Finally,the development trend,key technolo⁃gies and application prospects of clothing style recognition classification in fashion trend prediction and clothing collocation recommendation are analyzed and discussed.The study believes that in the future,it can develop in the direction of hu⁃man body-related feature extraction,multi-angle fashion trend prediction,and subjective evaluation of clothing recommen⁃dation.

关 键 词：深度学习 服装风格 卷积神经网络 特征提取 

分 类 号：TS941[轻工技术与工程—服装设计与工程] TP391.4[自动化与计算机技术—计算机应用技术]