基于弱监督学习面向主题的图文识别方法  

Subject-Oriented Text Recognition Method Based on Weakly Supervised Learning

作　　者：朱命冬 刘小杰 王鲜芳 张建霞 ZHU Mingdong;LIU XiaoJie;WANG Xianfang;ZHANG Jianxia(School of Computer Science and Technology,Henan Institute of Technology,Xinxiang 453003,China;Intelligent Industrial Big Data Application Engineering Technology Research Center of Xinxiang,Xinxiang 453003,China;School of Intelligent Engineering,Henan Institute of Technology,Xinxiang 453003,China)

机构地区：[1]河南工学院计算机科学与技术学院,河南新乡453003 [2]新乡市智能工业大数据应用工程技术研究中心,河南新乡453003 [3]河南工学院智能工程学院,河南新乡453003

出　　处：《河南工学院学报》2021年第6期29-32,40,共5页Journal of Henan Institute of Technology

基　　金：国家自然科学基金资助项目(61802116,62072157);河南省高等学校青年骨干教师培养计划(2020GGJS263);河南省科技攻关项目(202102210372,202102210168);河南工学院高层次人才科研启动基金(KQ1818)。

摘　　要：图文识别方法一直是计算机视觉领域的一个重要问题,随着深度学习技术的发展,这个问题成为学术界和产业界的研究热点。但现有的面向主题的数据集较少且易含有噪音,针对这个挑战,一种基于弱监督学习面向主题的图文识别方法被提出来,该方法结合区域卷积神经网络和长短记忆模型,通过弱监督学习方法和基于位置信息的全局连接层,实现对图片中主题文本信息准确连贯地识别。通过对图书封面数据集进行测试,并与现有相关方法进行了对比,验证了所提方法的准确性和高效性。The text recognition method has always been an important research problem in the field of computer vision.With the development of deep learning technology,this problem has become a research hotspot in academia and industry,and the corresponding research results have also been proposed.However,the existing subject-oriented data sets are small and easily contain noisy data.In response to this challenge,a subject-oriented text recognition method based on weakly supervised learning(STRW) is proposed,which combines RCNN and LSTM based on the attention mechanism.The model,through the weakly supervised learning method and the global connection layer based on location information,realizes the accurate and coherent recognition of the subject text information in the image.In the experiment,training and testing were conducted based the collected book cover data set.By comparing with the existing related methods the accuracy and efficiency of the STRW method is verified.

关 键 词：图文识别 弱监督学习 面向主题 相似性计算 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]