基于迁移学习的卷积神经网络印刷汉字字体识别模型研究

Study on the Font Type Recognition Model of Printed Chinese Character Using Convolutional Neural Network Based on Transfer Learning

汉字字体识别是光学字符识别技术(Optical Character Recognition,OCR)中的重要组成部分,归属于模式识别领域。针对目前字体识别方法在多字体识别和特征提取方面存在的问题,本研究提出基于迁移学习方法的深度卷积神经网络模型的印刷汉字字体识别新模型。根据印刷汉字字体识别任务特点对Inception-v3模型结构进行修改,制作印刷汉字字体识别数据集,并通过迁移学习的方法对模型完成训练。实验结果显示,该方法的识别平均准确率为99.83%,与相关的卷积神经网络模型相比,该模型的特征提取能力更强,识别准确率更高。

Font type recognition of Chinese character is an important part of OCR(Optical Character Recognition)technology,which belongs to the field of pattern recognition.Aiming at the existing problems in current font recognition methods used in multi-font recognition and feature extraction,a new model of Chinese character font recognition based on the method of transfer learning and the model of deep convolution neural network was proposed in this study.According to the characteristics of Chinese character font recognition task,the structure of the Inception-v3 model was modified and the Chinese character font recognition data set was made,and the model was trained by the method of transfer learning.The experimental results showed that the average recognition accuracy of this model is 99.83%.Compared with the related convolutional neural network model,this model has stronger feature extraction capability and higher recognition accuracy.