古籍汉字图像的可变形卷积网络检索模型  

Deformable convolutional network retrieval model for ancient Chinese character images

作　　者：田学东 王志红 左丽娜 TIAN Xuedong;WANG Zhihong;ZUO Lina(School of Cyber Security and Computer,Hebei University,Baoding,Hebei 071002,China)

机构地区：[1]河北大学网络空间安全与计算机学院,河北保定071002

出　　处：《中国科技论文》2020年第4期461-468,共8页China Sciencepaper

基　　金：河北省高等学校科学技术研究重点项目(ZD2017208)。

摘　　要：针对现有图像检索技术应用于古籍汉字图像时效果欠佳的问题,在传统卷积神经网络(convolutional neural networks,CNN)的基础上,引入可变形卷积构建适用于古籍汉字图像的CNN模型。首先,利用该模型对古籍汉字图像数据集进行特征提取;然后,利用主成分分析法对特征进行降维;最后,度量查询图像和候选图像的余弦相似度,排序并返回结果。所提出的检索方法在古籍汉字图像数据集上的平均精度均值达到70.42%,平均检索用时为3.15 s。实验结果表明,该模型能够有效地提取古籍汉字图像的特征,提高了检索方法的准确率,在古籍汉字图像检索领域具有一定优势。Aiming at the problem that the existing image retrieval technology was not effective when applied to ancient Chinese characters,a deformable convolution is introduced to build a convolutional neural networks(CNN)model suitable for ancient Chinese character images.Firstly,the features of ancient Chinese character image dataset were extracted with the CNN model.Then,the dimension reduction was realized by the principal component analysis.Finally,the cosine similarity between the query image and the candidate images was calculated and the hit images were sorted for users.The mean average precision of the proposed method is 70.42%,and the average retrieval time is 3.15 s.The experimental results show that the model can effectively extract the features of ancient Chinese character images and improve the precision of the retrieval method,which has advantages in the field of ancient Chinese character image retrieval.

关 键 词：计算机图像处理 古籍汉字图像检索 卷积神经网络 可变形卷积 主成分分析 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]