Journal of Software:2020.31(2):531-543

(燕山大学 信息科学与工程学院, 河北 秦皇岛 066004;河北省信息传输与信号处理重点实验室(燕山大学), 河北 秦皇岛 066004;北京交通大学 信息科学研究所, 北京 100044)
Image Hierarchical Classification Based on Semantic Label Generation and Partial Order Structure
GU Guang-Hua,CAO Yu-Yao,LI Gang,ZHAO Yao
(School of Information Science and Engineering, Yanshan University, Qinhuangdao 066004, China;Hebei Key Laboratory of Information Transmission and Signal Processing(Yanshan University), Qinhuangdao 066004, China;Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China)
Chart / table
Similar Articles
Article :Browse 177   Download 69
Received:October 25, 2017    Revised:December 31, 2017
> 中文摘要: 智能电子设备和互联网的普及,使得图像数据爆炸性膨胀.为了有效管理复杂图像资源,提出一种基于加权语义邻近集和形式概念偏序结构的图像层级分类方法.首先,根据图像语义相关分数,对不同程度语义设定自适应权系数,从训练图库中构建加权语义邻近集,通过对语义邻近集中图像的词频分布进行判决,自动生成图像的多个语义标签;然后,以每幅图像为对象,以每幅图像自动生成的语义标签为属性,构建形式背景,通过偏序结构算法对复杂图像集进行有效的层级分类.该方法可以得到图像库中图像之间明确的结构关系和图像类别之间的从属关系,为复杂图像大数据进行层级分类管理提供了有效的思路.对Corel5k、EspGame和Iaprtc12这3个数据库进行了图像标注实验,证明了标注的语义完整性和主要语义的准确性;并对Corel5k数据库进行了图像的层级分类实验,结果表明,层级分类效果显著.
Abstract:The popularity of smart electronic devices and the Internet makes the image data explode. In order to effectively manage the complex image resources, this study proposed an image hierarchical classification method based on a weighted semantic neighborhood set and formal concept partial order structure. Firstly, a weighted coefficient on different semantics is adaptively designed by the image semantic correlation scores, and a weighted semantic neighborhood (WSN) is constructed from the training sets. The semantic labels of the images are automatically generated by judging the word frequency distribution of the images in the semantic neighborhood set. Then, the context is built by taking the images as the objects and the semantic labels as the attributes. This study also proposed an efficient hierarchical classification method for complex image dataset based on the partial order structure. The hierarchical classification method can get the explicit structure relation and the subordinate relationship between the image categories, which provides an effective idea for the hierarchical classification management of the complex images of large data. Three datasets Corel5k, EspGame, and Iaprtc12 were labeled by the WSN method. The label result proved the integrity of the image semantics and the accuracy of the main semantics. Further, the Corel5k dataset was performed by the hierarchical classification method. The experimental results showed the significant performance of the hierarchical classification.
文章编号:     中图分类号:TP391    文献标志码:
基金项目:国家自然科学基金(61303128);河北省自然科学基金(F2017203169,F2018203239);河北省高等学校科学研究重点项目(ZD2017080);河北省留学回国人员科技活动项目(CL201621) 国家自然科学基金(61303128);河北省自然科学基金(F2017203169,F2018203239);河北省高等学校科学研究重点项目(ZD2017080);河北省留学回国人员科技活动项目(CL201621)
Foundation items:National Natural Science Foundation of China (61303128); Natural Science Foundation of Hebei Province (F2017203169, F2018203239); Key Research Project of Higher Education of Hebei Province (ZD2017080); Science and Technology Foundation for Returned Overseas Talents of Hebei Province (CL201621)
Reference text:


GU Guang-Hua,CAO Yu-Yao,LI Gang,ZHAO Yao.Image Hierarchical Classification Based on Semantic Label Generation and Partial Order Structure.Journal of Software,2020,31(2):531-543