摘要
针对基于字符分割的中文手写识别方法存在字符分割准确率影响识别准确率和速度的问题,文中设计了一种基于卷积循环神经网络(Convolutional Recurrent Neural Network,CRNN)的中文手写识别方法,方法将特征提取、序列预测、序列对齐算法集成到同一网络,实现端到端(End-to-End)的训练和识别。模型仅需输入待识别中文手写图像,中文字符免分割,即可输出识别结果,识别的准确率(Accuracy)相较基于结构特征点字符分割识别算法提高了2.29%,同时少了识别的时间。
Aiming at the problem that the character segmentation accuracy of Chinese handwriting recognition method which based on character segmentation has an influence on recognition accuracy and speed,a Chinese handwriting recognition method based on Convolutional Recurrent Neural Network(CRNN)is designed in this paper.Feature extraction,sequence prediction,and sequence alignment algorithms are integrated into the same network for end-to-end training and recognition.The model only needs to input the Chinese handwritten image to be recognized,and the recognition result with non-segmentation can be output.The Accuracy of recognition is 2.29%,which is higher than the character segmentation recognition algorithm based on structural feature points,and the recognition time is reduced.
作者
石鑫
董宝良
王俊丰
SHI Xin;DONG Bao-liang;WANG Jun-feng(North China Institute of Computer Technology,Beijing 100083,China)
出处
《信息技术》
2019年第11期141-144,150,共5页
Information Technology
关键词
中文手写识别
CRNN
端到端
免分割
Chinese handwriting recognition
CRNN
end-to-end
non-segmentation
