Research On Handwritten Chinese Character Recognition Based On Deep Learning

Along with the advance of human life,Human-Computer Interaction has become one of the most important technologies,in all kinds of daily information processing.Moreover,in today's era,Deep Learning has solved many problems of Human-Computer Interaction for us.It can automatically mine and process the potential connection of data,which avoids the disadvantages of traditional methods.In the fields of handwritten notes collection,examination paper information collection,real-time reading for the blind,and handwritten ancient literature collection,we can apply handwritten Chinese character recognition technology,thus making the application in these fields becoming more practical.Chinese characters record information in the development of Chinese culture.The linkage benefits and identification problems of Chinese characters make the research content become one of the hottest issues in the field of Human Computer Interaction.Therefore,it has a natural advantage to adopt the Human Computer Interaction-Deep Learning technology in the application of handwritten Chinese character recognition.This paper proposes a method of handwritten Chinese character recognition based on Deep Learning.This system can recognize the complete handwritten Chinese character information and has a high recognition rate on the similar characters.The main steps required for recognition of handwritten Chinese characters are image preprocessing,feature extraction and classifier recognition.In this paper,the main work is as follows:Preprocessing of handwritten Chinese characters.In order to obtain effective and accurate recognition results,we should prepare the training-testing data and the image to be identified.Firstly,the trichromatic channel image is converted into gray and binary image to reduce the amount of original data.Then,this paper proposes an improved method,which uses histogram projection,function derivation and threshold segmentation to segment a text into a single word.And then,this paper reserves the image on the basis of the preliminary single word image.In this paper,the morphological image transformation method and the improved method composed of 14 templates(3*3)are used for font smoothing and refinement.Finally,in order to match the recognition model,the final image is normalized by bicubic algorithm.Apply the Deep Learning technology to the feature extraction and classification recognition of handwritten Chinese characters.Using HWDB,a Chinese handwriting sample,library from the institute of automation,Chinese academy of sciences(CASIA)serve as training and testing images.In this paper,the data and corresponding labels are pre-established in batch,in order to complete the preparation of dataset.After analyzing many experiments of different schemes,the optimal model suitable for this application scenario is finally found.The construction of software platform and analysis of experimental results.Firstly,Machine learning is done on Linux,the Ubuntu desktop operating system is created and configured in the VMware Workstation,and GPU-Tensorflow is performed on the Ubuntu command line.Finally,a Human-computer interaction GUI interface based on Python is designed to realize the recognition of handwritten Chinese characters.And,The experimental results are evaluated on the basis of the pictures to be recognized and similar words similar words corresponding to each picture and also tne proposing for development of this system.