[1]牛少刚,张素兰,张继福.基于视觉注意力和 FCA 的古建筑图像语义完备[J].计算机技术与发展,2022,32(09):214-220.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 033]
　NIU Shao-gang,ZHANG Su-lan,ZHANG Ji-fu.Semantic Completion Annotation of Ancient Architecture Based on Visual Attention Mechanism and FCA[J].,2022,32(09):214-220.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 033]
点击复制

基于视觉注意力和 FCA 的古建筑图像语义完备()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年09期
页码:
214-220
栏目:
新型计算应用系统
出版日期:
2022-09-10

文章信息/Info

Title:
Semantic Completion Annotation of Ancient Architecture Based on Visual Attention Mechanism and FCA
文章编号:
1673-629X(2022)09-0214-07
作者:
牛少刚; 张素兰; 张继福
太原科技大学 计算机科学与技术学院,山西 太原 030024
Author(s):
NIU Shao-gang; ZHANG Su-lan; ZHANG Ji-fu
School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China
关键词:
古建筑图像; 标签完备; 卷积神经网络; 视觉注意力机制; 形式概念分析
Keywords:
ancient architecture images; tag completion; convolutional neural networks; visual attention mechanism; formal concept analy鄄sis
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 09. 033
摘要:
准确完备的古建筑图像语义不仅可提高古建筑图像检索效率,且能有效反映古建筑的历史文化信息。 针对不同古建筑图像轮廓特征明显不同且建筑 语义互相关联,为有效丰富古建筑图像语义,提出一种基于视觉注意力机制和形式概念分析( Formal Concept Analysis,FCA)的古建筑图像语义完备方法。 首先使用注意力算子网络和 VGG16 网络模型生成待标注古建筑图像注意力图,并通过 softmax 分类器进行分类,获取图像初始标签集;其次构造基于待标注图像初始标签及其近邻标签的概念格;然后,利用概念格上下文分析语义的特点,通过概念节点之间的相似度度量,获取待标注图像潜在的语义标签。 最后,在古建筑图像数据集上进行实验,结果验证了该方法能够有效地提高古建筑图像标注精度,丰富古建筑图像语义。
Abstract:
Accurate and complete ancient architectural image semantics can improve the efficiency of image retrieval and effectively reflect the historical? and cultural information. Aiming at the problem that the contour features of different ancient architectural images are different and the architectural semantics are interrelated,we propose a semantic completion method of ancient architectural images based on visual attention mechanism and FCA ( Formal Concept Analysis) ,which effectively enriches image semantics. Firstly,the attention operator network and VGG16 network model are used to generate the attention map of ancient architecture,and the initial label set of the image is obtained by softmax classifier. Secondly,we construct a concept lattice based on the initial label of the image to be labeled and its neighbor labels. Then we use the context of the concept lattice to analyze the semantic characteristics and obtain the latent semanticlabels of the image to be labeled by measuring the similarity between the concept nodes. Finally,experiments on the ancient architecture image datasets show that the proposed method can effectively complete the semantics of the ancient architecture image and improve the accuracy of image annotation.
更新日期/Last Update: 2022-09-10