Research Of Off-Line Handwritten Chinese Character Recognition System Based On Convolution Neural Network

Offline handwritten Chinese character recognition is a very important branch in the field of pattern recognition,and it is also a key technology in the process of computer intelligent interface implementation.Due to the large size of offline handwritten Chinese characters,the complicated font structure,similar Chinese characters,large writing arbitrariness and strong writing style,it has been widely concerned and studied in the field of pattern recognition.It is a hot and difficult point for many scholars to study The With the depth of learning in the field of machine learning gradually and rapid development,while in the pattern recognition part has also achieved excellent recognition performance.The advantage of convolution neural network in depth learning is that it can automatically obtain the characteristics of learning samples,which can avoid the problem of handwritten Chinese character extraction,but it is more difficult to train the samples.In this paper,convolution neural network is used to study the offline handwritten Chinese character recognition system,and the problem of training sample is optimized and improved.The main research work is as follows:（1）The feasibility of convolution neural network in off-line handwritten Chinese character recognition is verified by studying and analyzing the related technology of handwritten Chinese character recognition and the theory of convolution neural network.（2）Through the comparison of HCL2000 and CSAIA-HWDB two types of off-line handwritten Chinese character database,and the comparison between the system and the CSAIA-HWDB two-type off-line handwritten Chinese character database,The hardware part of the GPU architecture design,to achieve the GPU,CPU and SSD hard drive between the three data information interaction process,combined with Caffe software platform design,based on convolution neural network off-line handwritten Chinese character recognition system;（3）In the realization of the offline handwritten Chinese character recognition system,through the structure optimization of the convolution neural network,this paper proposes four improved network models from the aspects of the activation function of the convolution neural network,the dropout method and the elastic deformation of the data sample image.The experimental results show that the improved CNN₄ convolution neural network has improved the accuracy of Chinese character recognition by 2.5%in the process of offline handwritten Chinese character recognition.It is found by experiments that the improved convolution neural network improves the generalization ability,convergence speed and recognition accuracy of the network model in off-line handwriting recognition.