结合深度特征与多标记分类的图像语义标注  

Combining Deep Feature and Multi-label Classification for Semantic Image Annotation

作　　者：李志欣[1,2] 郑永哲 张灿龙 史忠植[3] Li Zhixin;Zheng Yongzhe;Zhang Canlong;Shi Zhongzhi(Guangxi Key Lab of Multi-source Information Mining and Security,Guangxi Normal University,Guilin 541004;Guangxi Collaborative Innovation Center of Multi-source Information Integration and Intelligent Processing,Guilin 541004;Key Laboratory of Intelligent Information Processing,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190)

机构地区：[1]广西师范大学广西多源信息挖掘与安全重点实验室,桂林541004 [2]广西区域多源信息集成与智能处理协同创新中心,桂林541004 [3]中国科学院计算技术研究所智能信息处理重点实验室,北京100190

出　　处：《计算机辅助设计与图形学学报》2018年第2期318-326,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61663004;61363035;61365009);国家"九七三"重点基础研究发展计划(2013CB329502);广西自然科学基金(2016GXNSFAA380146;2014GXNSFAA118368);广西多源信息挖掘与安全重点实验室主任基金(16-A-03-02)

摘　　要：为了缩减不同模态数据间的语义鸿沟,提出一种结合深度卷积神经网络和集成分类器链的多标记图像语义标注方法.该方法主要由生成式特征学习和判别式语义学习2个阶段构成.首先利用深度卷积神经网络学习图像的高层视觉特征;然后基于获取的视觉特征与图像的语义标记集训练集成分类器链,并学习视觉特征包含的语义信息;最后利用训练得到的模型对未知图像进行自动语义标注.在Corel5K和PASCAL VOC 2012图像数据集上的实验结果表明,与一些当前国际先进水平的方法相比,文中方法的鲁棒性更强,标注结果更精确.To bridge the semantic gap between data of difference modalities,a multi-label semantic image annotation approach combining deep CNN(convolutional neural network)and ECC(ensemble of classifier chains)is proposed.The approach is composed of two stages,i.e.generative feature learning and discriminative semantic learning.Deep CNN is utilized to extract high level visual features of images in the first step.Then ECC is trained based on the obtained visual features and image semantic label set.Meanwhile,semantic information in visual features is learned.Eventually,automatic annotation on unseen images can be done by the trained model.Experimental results on Corel5K and PASCAL VOC 2012 image datasets show that the proposed approach is more robust and more accurate than some state-of-the-art approaches.

关 键 词：图像语义标注 卷积神经网络 集成分类器链 深度特征 多标记分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]