Printed Chinese Character Recognition Based On Deep Learning

With the development of the Internet and the rise of deep learning,traditional opt ical character recognition(OCR)technology faces new challenges.After the analysis o f the available technologies of OCR and the application requirements,this paper desig n a method of Chinese character recognition based on deep learning and a software a pplication of Chinese character recognition based on deep learning.The first step is to generate large-scale printed Chinese character data sets.Throug h the technologies of image generation and image augment,this paper gets approximat ely 20 million Chinese character images.These images are 32x32 and 100x100 resolut ion and labeled for the deep learning model train.Based on the data sets build in this paper,this paper designs and selects 5 kinds of deep learning network models for sample training tuning and performance comparat ive analysis of printed Chinese character recognition.The size of the samples are 32x32 and 100x100.The models which used in the training are LeNet5,LeNet5-A,VGG7-A,ResNet-18 and ResNet-50.Based on the results of training,the best model ResN et-18 is integrated in the Chinese character recognition application software.Based on the deep learning model of printed Chinese character recognition in this paper,a Chinese character recognition system for ID cards was designed.For the im ages of the ID cards,there are many steps of functions was constructed which includ e image classification,image pre-processing,character layout analysis and character rec ognition.All of these make a highly recognition rate ID card identification system.