Research On Handwritten Chinese Text Recognition Algorithm Based On Neural Network

As an important carrier for human beings to perceive the world and communicate information,texts play an important role in people’s daily life.With the development of science and technology,the digitization of information is an inevitable trend,but handwritten texts still be seen everywhere.In the fast-paced social life,the demand for automatic recognition of handwritten texts is increasing.Handwritten Chinese character text recognition has always been attracted the attention of the research community.Although Chinese character text recognition has been developed for many years,the performance of traditional methods cannot meet the actual needs due to the large character set,the different writing styles,the overlapping and adhesion of characters,etc.and the task of recognizing handwritten Chinese character text is still very challenging.With the rapid development of deep learning technology,it provides more effective solutions in the field of computer vision.Therefore,this paper adopts deep learning technology to study the recognition of handwritten Chinese text.The main research work is as follows:To address the problems of handwritten Chinese character recognition,such as many kinds of characters,complex character structure,confusion of similar characters,and different handwriting styles,this paper proposes a handwritten Chinese character recognition method based on improved VGG16 network.Aiming at the problems that vgg16 recognition model consumes a lot of computing resources and does not make full use of hierarchical feature map,this paper proposes methods for improvement separately.By using the Ghost module to replace the operation of the ordinary convolutional layer,the amount of parameters and computation of the network is effectively reduced.In order to fully reuse the hierarchical features generated by different network layers,this paper proposes a feature reuse network,which effectively enhances the feature representation of handwritten images and improves the recognition performance.Aiming at the difficulty of character segmentation in handwritten Chinese text,this paper proposes a segmentation-free method for handwritten Chinese text recognition that integrates multi-level context information,Multi-Local and Global Context Network(MLGCNet).Since the traditional CRNN model uses LSTM to model the context of the feature sequence,this process cannot be processed in parallel and the long-term context relationship cannot be effectively capture when processing long sequences,resulting in the reduction of recognition effect.Therefore,this paper proposes a multi-level sequence modeling method integrating multi local context and global context to replace the sequence modeling process of LSTM.This method is easy to execute in parallel and can make full use of context information.The experimental results show that the recognition performance of MLGCNet has been improved compared with CRNN,and it can also achieve more competitive results in the case of low-resolution image.