基于卷积神经网络和概念格的图像语义完备标注  

Image Semantic Completion Annotation Based on CNN and Concept Lattice

作　　者：李雯莉 张素兰[1] 张继福[1] 胡立华[1] LI Wen-li;ZHANG Su-lan;ZHANG Ji-fu;HU Li-hua(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,太原030024

出　　处：《小型微型计算机系统》2020年第9期1979-1986,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61373099)资助。

摘　　要：如何有效地提取图像底层特征、分析高层语义的潜在语义关系,已成为图像标签完备标注亟待解决的问题之一.为有效完善图像语义标签,本文采用卷积神经元网络(CNN)和概念格,提出一种图像语义完备自动标注方法.首先构建自适应CNN网络,分割待标注图像并提取其特征,以此来获得近邻图像集与其一系列相对应的标签集合;然后利用概念格进行标签本身潜在的语义分析,有效地改善了标注效果,并保证了语义标注的完备性;最后利用投票的方式,得到最优语义标签.采用基准数据集Corel5k进行实验,验证了该方法能有效地丰富图像标签语义,提高标签召回率,并提高了图像语义检索效率.How to effectively extract and learn the underlying features of images and analyze the latent semantic relations of high-level semantics has become an urgent problem to be solved in image completion annotation.In order to effectively improve the image semantic label,this paper proposes a completion automatic annotation method for image semantics by using convolutional neural network and concept lattice.Firstly,an adaptive CNN network is constructed,and the image to be labeled is segmented and its features are extracted to obtain a set of neighboring image sets corresponding to a series of tags.Then the concept lattice is used to analyze the potential semantics of the tag itself,which effectively improves the labeling effect,and ensures the completeness of the semantic annotation;finally,using the voting method,get the optimal semantic label.Finally,experiments on benchmark dataset Corel5K,show that our method can effectively enrich the image label semantics,improve the label recall rate,and improve the image semantic retrieval efficiency.

关 键 词：图像完备标注 卷积神经网络(CNN) 概念格 语义扩展 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]