标题:A Dual-CNN Model for Multi-label Classification by Leveraging Co-occurrence Dependencies Between Labels
作者:Zhang, Peng-Fei; Wu, Hao-Yi; Xu, Xin-Shun
通讯作者:Xu, XinShun;Xu, XS
作者机构:[Zhang, Peng-Fei; Xu, Xin-Shun] Shandong Univ, Sch Comp Sci & Technol, Jinan, Shandong, Peoples R China.; [Wu, Hao-Yi] Univ Tasmania, Fac Sci Engn & 更多
会议名称:18th Pacific-Rim Conference on Multimedia (PCM)
会议日期:SEP 28-29, 2017
来源:ADVANCES IN MULTIMEDIA INFORMATION PROCESSING - PCM 2017, PT I
出版年:2018
卷:10735
页码:315-324
DOI:10.1007/978-3-319-77380-3_30
关键词:CNN; Multi-label; Classification; Label co-occurrence dependencies
摘要:In recent years, deep convolutional neural network (CNN) has demonstrated its great power in image classification. In real world, there are many images contain abundant contents so that they have multiple labels. Moreover, there are correlations between labels. Traditional deep methods for such data rarely take into account such correlations. In this paper, we propose a dual-CNN model, i.e., Dual-CNN model for Multi-Label classification (Dual-CNN-ML), which can make full use of the dependencies of labels to enhance classification performance. Specifically, we first obtain co-occurrence dependency matrix from training datasets; then, we merge the co-occurrence dependency matrix and image representation together; finally, we use the new representation to predict labels of samples. Extensive experiments on public benchmark datasets demonstrate that the proposed method obtains satisfying results and outperforms several state-of-the-art methods.
收录类别:CPCI-S;EI;SCOPUS
资源类型:会议论文;期刊论文
原文链接:https://www.scopus.com/inward/record.uri?eid=2-s2.0-85047480399&doi=10.1007%2f978-3-319-77380-3_30&partnerID=40&md5=a399b60bffced98f635a7714b660f280
TOP