摘要
为实现跨域情形下精确的服装图像检索,从关键区域识别技术和深度度量学习技术两个方面总结了最新研究进展,分析了现有研究中存在的问题。发现关键区域识别技术有助于服装关键特征的提取,可以有效的提升检索精度,但对具有相似特征不同类别的困难样本判别效果不佳,深度度量学习技术是解决这一问题的有效途径,利用不同损失函数的优化以及集成网络学习加强了服装特征的判别。最后通过实验结果对比,分析关键区域识别技术和深度度量学习技术跨域服装检索准确率,认为未来跨域服装图像检索准确率的提升主要依赖于服装关键特征提取和服装特征判别。
In order to achieve accurate clothing image retrieval in cross-domain situations,the latest research progress is summarized from two aspects of region recognition technology and deep metric learning technology,and the existing problems in existing research are analyzed.It is found that the critical region recognition technology is helpful for the extraction of critical features of clothing,which can effectively improve the retrieval accuracy,but the effect of discriminating difficult samples with similar features and different categories is not good.Deep metric learning technology is an effective way to solve this problem.The optimization of different loss functions and integrated network learning strengthen the discrimination of clothing features;finally,through the comparison of experimental results,the accuracy of cross-do-main clothing retrieval by critical region recognition technology and deep metric learning technology is analyzed,and it is believed that the accuracy of cross-domain clothing image retrieval will be improved in the future.It mainly depends on the critical feature extraction of clothing and the identification of clothing features.
作者
杨迪
陈宁
Yang Di;Chen Ning(School of Electronics and Information,Xi'an Polytechnic University,Xi'an 710048,China)
出处
《国外电子测量技术》
北大核心
2021年第11期24-34,共11页
Foreign Electronic Measurement Technology
关键词
跨域服装图像检索
关键区域识别
深度度量学习
深度学习
cross-domain clothing retrieval
critical region recognition
deep metric learning
deep learning
作者简介
杨迪,硕士研究生,主要研究方向为深度学习和服装图像处理技术。E-mail:1564773884@qq.com;陈宁,副教授,硕士生导师,主要研究方向为分布式网络人工智能、图像处理与识别、计算机控制系统等。E-mail:chennvictor@gmail.com。
