摘要
提出一种融合自适应图像增强的IE-AlexNet少数民族文字图像识别方法。以阿拉伯文字为例,计算各场景的图像噪声阈值,利用六种方法实现图像增强,然后构建一种改进AlexNet的神经网络模型,实现阿拉伯文字识别。实验结果表明,所提方法能有效检测复杂环境下的少数民族文字,其F1值为0.9559,准确率为0.9553,均优于其他模型。同时,对比实验突显图像增强的有效性,具有一定应用价值。
In this paper,an IE-AlexNet minority character image recognition method with adaptive image enhancement is proposed.Taking Arabic characters as an example,the image noise threshold of each scene is calculated,and six methods are used to realize image enhancement.An improved AlexNet neural network model is constructed to realize Arabic character recognition.Experimental results show that the proposed method can effectively detect minority characters in complex environments with an F1-score of 0.9559 and an accuracy rate of 0.9553,both of which are better than other models.The contrast experiment highlights the effectiveness of image enhancement,which has certain application value.
作者
杨秀璋
周既松
武帅
陈登建
刘建义
宋籍文
Yang Xiuzhang;Zhou Jisong;Wu Shuai;Chen Dengjian;Liu Jianyi;Song Jiwen(School of Information of Guizhou University of Finance and Economics,Guiyang,Guizhou 550025,China;Lianshui County Finance Bureau;Guizhou Expressway Group Co.,Ltd.)
出处
《计算机时代》
2022年第11期15-20,共6页
Computer Era
基金
贵州省科技计划项目(黔科合基础[2019]1041,黔科合基础[2020]1Y279)
贵州省教育厅青年科技人才成长项目(黔教合KY字[2021]135)
贵州财经大学2021年度校级项目(No.2021KYQN03)。
关键词
少数民族文字
深度学习
图像识别
数字人文
minority characters
deep learning
image recognition
digital humanities
