基于层次匹配的维吾尔文关键词图像检索  

Uyghur keyword image retrieval based on hierarchical matching

作　　者：宋志平 朱亚俐[1] 吾尔尼沙·买买提 徐学斌 库尔班·吾布力[1,2] SONG Zhi-ping;ZHU Ya-li;Hornisa·Mamat;XU Xue-bin;Kurban·Ubul(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;Key Laboratory of Multilingual Information Technology,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046 [2]新疆大学多语种信息技术重点实验室,新疆乌鲁木齐830046

出　　处：《计算机工程与设计》2022年第12期3461-3467,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61563052、61862061、61363064);新疆大学博士科研启动基金项目(BS150262)。

摘　　要：为提高维吾尔文文档图像检索精度,提出基于灰度共生矩阵(GLCM)与卷积神经网络的关键词图像分层检索算法。在浅层检索阶段对切分后的单词图像进行分块处理,计算每个子块图像的灰度共生矩阵特征参数,将各个子块特征进行串联融合,对单词数据库进行浅层检索,过滤掉部分无关单词图像后形成候选单词库;在浅层检索的基础上进行深层检索,使用VGG16网络提取单词图像更深层次的空间域特征;使用网络提取的特征对候选图像库进行二次深层检索得到最终的检索结果。实验结果表明,检索的平均准确率和召回率分别为94.15%、82.03%,验证了该方法在维吾尔文文档图像检索中的有效性。To improve the retrieval accuracy of Uyghur document image,a hierarchical keyword image retrieval algorithm based on GLCM and convolutional neural network was proposed.The segmented word image was segmented in the shallow retrieval stage,and the characteristic parameters of gray level co-occurrence matrix of each sub block image were calculated.The sub block features were fused in series,and the word database was searched superficially to filter out some irrelevant word images to form a candidate word database.Deep retrieval was carried out on the basis of shallow retrieval,and VGG16 network was used to extract deeper spatial features of word images.The features extracted from the network were used for the secondary deep retrieval of the candidate image database to obtain the final retrieval results.Experimental results show that the average accuracy and recall rate are 94.15%and 82.03%respectively,which verifies the effectiveness of the method in Uyghur document image retrieval.

关 键 词：维吾尔文 关键词检索 VGG16 灰度共生矩阵 单词切分 浅层检索 深层检索 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]