The Research Of Offline Handwritten Chinese Character Recognition Based On Deep Learning

Chinese characters are the most widely used words in the world.And Chinese character recognition has important application scenarios in the fields of disabled people's accessible reading,automatic document entry,mail sorting,bank bill processing,and identity recognition.There are a number of Chinese characters and the handwriting styles are different.Apart from that,there are a large number of similar characters in Chinese characters.So,the accuracy remains low in offline handwritten Chinese character recognition.In recent years,deep learning has developed rapidly,and has achieved good results in the fields of pattern recognition,natural language processing,and speech recognition.Therefore,this thesis studies offline handwritten Chinese character recognition based on the deep learning method.For the large classification problem of Chinese character recognition,the convolutional neural network in deep learning is used to recognize the most commonly used first-class 3755 Chinese characters collected in GB2312-80 standard.A typical convolutional neural network is an end-to-end structure,which directly uses the original image as its input,but it can't learn the relevant domain knowledge.In the conventional convolution operation,all channels in the corresponding area of the image are considered simultaneously,which inevitably increases the redundancy of the network.In this thesis,the eight-direction gradient feature of the image is used as the input of the convolutional neural network,and the depthwise separable convolution method is used to convolute the image.Finally,multiple groups of convolutional neural network are designed for experiments.The experimental results on the CASIA-HWDB dataset show that the eight-direction gradient feature input and the depthwise separable convolution can significantly improve the recognition effect of Chinese characters,and finally achieve an accuracy of 95.86%.For the problem that it is difficult to recognize similar characters in offline handwritten Chinese character recognition,this thesis improves it from two aspects.The first method combines convolutional neural network with center loss to recognize similar handwritten Chinese characters.The central loss function in metric learning is introduced to the convolutional neural network,and the cross-entropy loss and the center loss are used as the joint loss of the convolutional neural network.So the model can learn more discriminative features which can reduce the distance between same samples and increase the distance between different types of samples.The second method combines convolutional neural network with support vector machine to recognize similar handwritten Chinese characters.The convolutional neural network is regarded as a feature extractor,and the feature vectors of the convolutional neural network are used to train the support vector machine classifier to recognize similar handwritten Chinese characters.The experimental results show that the average recognition accuracy can be improved by 2.68% and 1.59% by using joint loss function and convolutional neural network plus support vector machine compared with using convolutional neural network alone.