Research And Implementation Of Chinese Character Font Recognition Technology Based On Deep Learning

With the development of the Internet and the rise of deep learning technology,traditional optical character recognition(OCR)faces new challenges in terms of layout recovery.We not only need to restore the text content on the target image,but also need to recognize the format of its Chinese font.However,there is a huge variability in Chinese fonts,and it is difficult for OCR to achieve convincing recognition accuracy for such versatile differences.Deep learning has made gratifying achievements in the fields of image recognition,natural language processing and speech recognition.The field of Chinese character font recognition has many commonalities with existing deep learning applications,and applies deep learning technology to the field of Chinese character font recognition.It can not only solve the problem of recognition of Chinese fonts,but also broaden the application field of deep learning.This paper mainly uses TensorFlow deep learning framework,VGG-Net convolutional neural network structure and Flask back-end Web framework to build a Chinese character font recognition system based on deep learning.The data set used in this experiment is a Chinese character font image extracted from the ttf file.Each font contains 3500 common word training set images and 712 unusual word test set images.In this paper,the Chinese character font recognition model is built for the two fonts of Kai Script and Li Script,and the performance is tested.The overall recognition accuracy rate is 99.7%.Compared with the Chinese character font recognition technology based on the empirical mode or the global texture feature,the accuracy rate has been greatly improved.After that,the model was transferred to the recognition of nine Chinese characters such as Kai Script,Li Script,Xing Script,Cao Script,Ou Script,Zhao Script,Weibei Script,Yan Script and Liu Script,and achieved 98.2%recognition accuracy on the test set.At the same time,this paper builds a Web system that can recognize the font image in real time by using the trained model.After testing,the system can recognize the incoming Chinese font image in real time and pass the recognized result back to the front page to display.