作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 开发研究与工程应用 • 上一篇    下一篇

基于深度卷积神经网络的服装图像分类检索算法

厉智1,孙玉宝1,2,王枫1,刘青山1,2   

  1. (1.南京信息工程大学 信息与控制学院 江苏省大数据分析技术重点实验室,南京 210044; 2.江苏省大气环境与装备技术协同创新中心,南京 210044)
  • 收稿日期:2015-10-29 出版日期:2016-11-15 发布日期:2016-11-15
  • 作者简介:厉智(1990—),男,硕士研究生,主研方向为深度学习、图像分类与检索;孙玉宝,副教授、CCF会员;王枫,硕士研究生;刘青山(通讯作者),教授、博士生导师、CCF会员。
  • 基金资助:
    国家自然科学基金(61672292,61300162,61272223);江苏省自然科学基金(BK20131003)。

Clothing Image Classification and Retrieval Algorithm Based on Deep Convolutional Neural Network

LI Zhi  1,SUN Yubao  1,2,WANG Feng  1,LIU Qingshan  1,2   

  1. (1.Jiangsu Province Key Laboratory of Big Data Analysis Technology,School of Information and Control, Nanjing University of Information Science and Technology,Nanjing 210044,China;2.Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment Technology,Nanjing 210044,China)
  • Received:2015-10-29 Online:2016-11-15 Published:2016-11-15

摘要: 针对利用深度学习的服装图像检索算法分类精度较低的问题,提出基于深度卷积神经网络的改进服装图像分类检索算法,并建立一个具有16种属性、10万量级的网络服装图像数据库B_DAT Clothing。依据服装图像的复杂性,采用深度卷积神经网络从B_DAT Clothing数据库中自动学习服装的属性特征并建立哈希索引,进而构建基于服装属性的检索模型,实现服装图像的高效分类和快速检索。实验结果表明,与传统视觉特征分类算法相比,该算法在服装分类上的准确率更高,属性检索效果更好。

关键词: 服装属性, 卷积神经网络, 属性检索, 分类排序, 哈希索引, 服装数据库

Abstract: Aiming at the problem that clothing image retrieval algorithm based on deep learning has low classification accuracy,this paper proposes an improved clothing image classification and retrieval algorithm based on Deep Convolutional Neural Network(DCNN).A clothing image database that contains 100 000 images with 16 attributes,called B_DAT Clothing,is established.Duo to the complex performance of clothing images,it uses DCNN to learn the features adaptively from the B_DAT Clothing Database,design the hash index of CNN’s features for building an efficient attribute-based retrieval model,and realize efficient classification and quick retrieval of Clothing images.Experimental results show that the algorithm can achieve better performance in terms of classification and retrieval than the traditional visual feature classification algorithms.

Key words: clothing attributes, Convolutional Neural Network(CNN), attribute retrieval, classification sorting, hash index, clothing database

中图分类号: