基于关键词图像识别的维-哈语文档图像分类  

Uyghur-Kazakh Document Image Classification Based on Keyword Image Recognition

作　　者：沙尔旦尔·帕尔哈提 木塔力甫·沙塔尔 阿力木江·亚森 阿布都热合曼·卡的尔 SARDAR Parhat;MUTALLIP Sattar;ALIMJAN Yasin;ABDURAHMAN Kadir(College of Information Management,Xinjiang University of Finance and Economics,Urumqi Xinjiang 830012,China)

机构地区：[1]新疆财经大学信息管理学院,新疆乌鲁木齐830012

出　　处：《计算机仿真》2025年第2期235-242,329,共9页Computer Simulation

基　　金：国家自然科学基金(61662073,62241208);新疆财经大学校级科研基金项目(2022XGC022,2022XGC049)。

摘　　要：对分类多个领域文档图像而言,识别并提取其关键词是不可或缺的。维-哈语等低资源语言文档图像分类研究匮乏,以往的维语文字图像识别方法不区分内容所属领域,用传统的文字图像识别框架,易受分辨率变化的影响,而哈语文字图像识别研究不足。针对以上问题,提出基于合成图像的维-哈语关键词图像识别方法,引入8种不同字体及弯曲、噪声等效果,使图像数据覆盖多种实际场景特征,并设计深层卷积神经网络,构建维-哈语关键词图像识别模型,获得了97.5%的准确率。进而,将文本分类与关键词图像识别模型相结合,进行了维-哈语文档图像分类,并获得了71%的准确率。实验表明,通过用关键词图像识别方法可以较好地实现维-哈语文档图像的自动分类。For classifying document images from multiple domains,identifying and extracting their keywords is essential.There is a lack of research on image classification of low resource language documents such as Uyghur and Kazakh.Previous Uyghur text image recognition methods do not distinguish the field of content and use traditional text image recognition frameworks,which are easily affected by resolution changes.However,research on Kazakh text image recognition is insufficient.For this problem,a Uyghur-Kazakh keyword image recognition method based on synthetic images is proposed.Eight different fonts,distortion,noise and other effects are introduced to make the image data cover a variety of actual scene features.A deep convolutional neural network is designed to construct the Uyghur-Kazakh keyword image recognition model,and the accuracy rate of 97.5%is obtained.Then,by combining keyword image recognition model with text classification model,the automatic classification of Uyghur and Kazakh document images is realized,and the accuracy rate of 71%is obtained.The experiment shows that the automatic classification of Uyghur and Kazakh document images can be achieved by using keyword image recognition method.

关 键 词：维-哈语 关键词图像识别 卷积神经网络 文档图像分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]