Handwritten English Handwriting Identification And Document Recognition Using CNN

With the development of artificial intelligence technology,it has been widely used in the field of computer vision and has become a hot topic.Its application has got wide attention in the field of English handwriting identification and handwritten English document recognition.In this paper,a CC-VGG network model is proposed to realize the English handwriting identification.Besides,an improved VGG-16 model is used to recognize the handwritten English documents,and perfect effects can be achieved.First,the research background and status are introduced in this thesis.The basic neural network theory is discussed,as well as the feedforward neural network,convolutional neural network(CNN),and the commonly used network models.Second,a CC-VGG network model is proposed based on the traditional VGG-16 model.Using the composite convolution layer to replace the partial convolution layer,the handwritten English identification is realized.Moreover,a handwritten English handwriting image data set EI130,which contains 130 categories and 26,000 pictures,was created.Using the proposed model on the public CVL and ICDAR2013 datasets,experimental results demonstrate that the average correct rate reaches 92.7% and86.9%,respectively,which is higher than that of the existing algorithms.Besides,the proposed model also achieves high accuracy on the established EI130 data set,which demonstrates its effectiveness and robustness.In addition,this thesis proposes an unknown handwriting identification criterion,and demonstrates its effectiveness with the simulation experiments.Besides,a handwritten English document recognition method is proposed on the base of the improved VGG-16 model.On the established English handwritten data set,the automatic recognition of the handwritten English document is realized using the constructed CNN model.Experimental results show that the proposed method can effectively realize the automatic recognition of handwritten English document and relative punctuation,and the recognition accuracy is above 99%.Moreover,the algorithm is robust to illumination changes,simple geometric deformation,and additional noise.Finally,the work in this thesis is summarized,the advantages and disadvantages of the proposed methods are analyzed,and the research direction in the future is expected and determined.