摘要
针对文本识别中存在的光照不均匀、字符质量低等问题,提出一种图像增强算法和卷积循环神经网络字符识别模型。图像增强算法使用考虑局部信息的改进色调映射函数增加暗区域文字的可见度。通过背景估计和对比度补偿的方法解决图像光照不均匀问题,使用连通域的方法对图像中的文字定位。基于文字区域搭建卷积和循环深度神经网络模型,以图像内整个字符串作为识别目标。采集30幅光照不均匀图像进行实验验证,结果表明该模型在该场景下的文字识别准确率为98.29%。
Aiming at the problem of uneven illumination and low character quality in text recognition,this paper proposes an image enhancement algorithm and a convolutional recurrent neural network character recognition model.In the image enhancement algorithm,the improved tone mapping function considering local information was used to increase the visibility of text in the dark region.Through the method of background estimation and contrast compensation to solve the problem of uneven illumination,we used the connected domain method to locate the text in the image.The convolutional and cyclic depth neural network model was built based on the text region,and the whole string in the image was taken as the recognition target.30 images of uneven illumination are collected for experimental verification,and the results show that the text recognition accuracy of the model in this scenario is 98.29%.
作者
何鎏一
杨国为
He Liuyi;Yang Guowei(School of Electronic Information,Qingdao University,Qingdao 266071,Shandong,China;School of Information Engineering,Nanjing Audit University,Nanjing 211815,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2020年第6期184-190,217,共8页
Computer Applications and Software
基金
国家自然科学基金面上项目(61772277)。