摘要
手写体识别一直是OCR领域的一个热点与难点,随着深度学习快速发展,在OCR领域取得不错的成果。论文设计了一种基于卷积循环神经网络(Convolutional Recurrent Neural Network,CRNN)和最大熵CTC(EnCTC)损失函数进行离线手写体英文识别方法,通过空间转换网络(STN)对数据样本进行几何转换,通过CNN网络提取文字图像特征序列,利用多层BiLSTM网络来学习特征序列的上下文信息,最后使用EnCTC损失函数来进行转录解码,在整个结构上实现了端到端(end-to-end)的识别,不需要对单词进行分开识别,通过对比实验来证明论文算法的有效性。
Handwriting recognition has always been a hot and difficult point in the field of OCR.With the rapid development of in-depth learning,good results have been achieved in the field of OCR.In this paper,an off-line handwritten English recognition method based on improved CRNN network is designed.Geometric transformation of data is set through STN,the feature sequence of text image is extracted through improved CNN network,and the multi-layer is used BiLSTM network is used to learn the context information of feature sequences.Finally,this paper uses the EnCTC loss function to decode transcription.In the whole structure,the end-to-end recognition is realized without separate words.The effectiveness of this method is proved by comparative experiments.
作者
朱世闻
ZHU Shiwen(College of Automation&College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210000)
出处
《计算机与数字工程》
2022年第5期1093-1097,共5页
Computer & Digital Engineering
