Covered Off-line Handwritten Chinese Character Recognition Based On Convolutional Neural Network

Character is the origin of Chinese culture and the unique culture of the Chinese nation.With the rapid development of science and technology and information technology,handwritten Chinese character recognition has been widely used in bill identification,email recognition,automated teaching and other fields,which has become an important subject in computer vision area.The accuracy rate of handwritten Chinese character is difficult to break through the bottleneck due to its feature of huge number of strokes,complex combination of diverse,high similarity,poor handwriting styles and so on.While there are many covered Chinese characters in real life,the difficulty of recognition is greatly enhanced.This paper studies the covered handwritten Chinese character recognition,the main work contents and innovation results are as follows:The convolutional neural network was used to identify the covered handwritten Chinese characters.The effects of the number of convolution kernels,the size of convolution kernels,the learning rate,and the number of batch samples on the recognition accuracy were investigated.The experimental results show that the number of convolution kernels is set to 64,128,256,the convolution kernel size is 5×5,the learning rate is set to 0.1,and the batch training sample is 64.In addition,increasing the number of convolutional network layers can improve accuracy.At the same time,the convolutional neural network is used as a feature extractor combined with traditional machine learning classifiers such as support vector machines,decision trees,and naive Bayes classifiers to further improve the recognition accuracy.Among them,support vector machine classification performs best for recognition accuracy rate which is 92.17%.The method of model fusion was introduced to further improve the recognition accuracy.This paper adopts three kinds of formal model fusion methods: parallel fusion based on output layer features,serial fusion based on fully connected layer features,and serial fusion based on output layer features.Experimental results show that the accuracy of experimental results of the three fusion methods is significant.The promotion,which based on the output layer feature of the serial fusion method works best.A sparse encoder was introduced as a feature extractor for the covered handwritten Chinese characters.The effects of the number of hidden layers and the number of different neurons in each hidden layer on the recognition accuracy were investigated.The experimental results showed that the number of neurons in each layer was increased and the number of neurons increased.The number of hidden layers can improve the recognition rate of handwritten Chinese characters.At the same time,compared with the traditional image feature method Gabor,Sobel,the result is that the sparse encoder works best.This paper combines deep learning technology with traditional methods and applies it to the covered handwritten Chinese character datasets.The recognition accuracy rate has been significantly improved.