Handwritten Chinese Character Recognition Based On PSO And BP Neural Network

Research on Chinese character recognition is one of the hot topics of pattern recognition and in the recent years it is gaining world-wide attention through successful implementation of recognition methods.In this thesis,a study of recognition of handwritten Chinese characters in offline mode is proposed.Considering the characteristics of Chinese character recognition process,a handwritten offline recognition model based on BP neural network is established,and its feasibility is verified by training and testing.Initially,the handwritten Chinese character sample image is selected to start preprocessing.Afterwards,the feature vector is extracted from the processed digital image in the feature extraction process.In the subsequent stages,using the advantages of neural networks,and considering the related features such as self-adaptation,self-learning and fault tolerance,the classification and recognition model of random sample data is designed.The developed recognition model improved to obtain higher recognition rate of handwritten Chinese characters and optimizes the process of Chinese character classification and recognition.This thesis focuses on three main aspects of Chinese character recognition.Firstly,the basic process of offline handwritten Chinese character recognition and the basic algorithm model of each process are summarized.The overall process includes image acquisition,image preprocessing,feature extraction,and recognition classification model construction.The sample images analyzed in this work are obtained from the SCUT-IRAC HCCLIB sample library.The image preprocessing step is divided into five processes:binarization,smooth denoising,character segmentation,normalization and refinement.Binarization confirms no blank points in the handwritten character strokes.The character strokes basically retain the basic features of the original characters.The smooth denoising process removes the isolated noise black points in the previous image.The character segmentation separates the images into single Chinese characters for simulation training recognition.The normalization process adjusts the change of the size and position of a single Chinese character.The refinement process removes the edge contour of the Chinese character and retains only the most basic shape information.The feature extraction process uses the grid direction statistical feature method,and the extracted feature vector is used as input for classification and recognition model simulation verification.Then the recognition classification model is designed based on BP neural network.Secondly,for accomplishing the process of Chinese character recognition and classification,this thesis introduces the principle,implementation and characteristics of BP algorithm and PSO algorithm.The BP algorithm is applied as a local optimization algorithm,when the BP neural network is used to classify handwritten Chinese characters.This tends to fall into local minimum points during the network training phase,and the convergence speed is slow,which leads to the reduction of classifier recognition rate.To overcome this problem,an algorithm based on PSO algorithm to optimize BP neural network(PSO-BPNN)is presented in this thesis,which establishes a Chinese character handwritten offline classification recognition model.Thirdly,the sample image in the database is used to train and test the effectiveness of the above PSO-BPNN algorithm.The simulation environment is developed using MATLAB.Initially,a certain number of sample images are selected for training in the early recognition classification model after preprocessing and feature extraction,and then the remaining sample images are used for the test verification of the model.Afterwards,the training simulation of a large number of sample images is carried out,and the recognition effects of the original BP algorithm and PSO-BPNN algorithm are compared.The verification shows that the optimized classification model has higher prediction accuracy and recognition ability.