Research On Character Image Recognition Based On Convolution Neural Network

With the continuous development of the Internet,all walks of life have entered the era of digital informatization.In order to improve the informatization level of various industries and use modern information technology to carry out efficient and convenient service work,the recognition of pictures becomes an urgent need.In this paper,the convolution neural network is used to identify the character type images.The sample images used are 23974,and 22,000 images are randomly selected as the training set samples.The remaining 1974 pictures are model for verification and identification of verification collection.First,we use the convolution neural network algorithm to process these images,which include the greyscale of the picture,binarization,de-noising and segmentation of images.Then,the sample images of the training set after processing are trained by the convolution neural network,which is the process of minimizing the loss function,searching for the optimal parameters,and verifying the 1974 images in the validation set.Finally,this paper analyzed the loss value and accuracy of the recognition results,and compared the results with the recognition results of an OCR software,and found that the recognition effect of the convolution neural network algorithm was better than the recognition effect of the OCR software from the perspective of recognition accuracy.In addition,the accuracy rate of the recognition of the convolution neural network was carried out with the accuracy rate of the OCR software recognition,which was verified by the statistical Angle,and when the significance level was 0.05,the accuracy rate of the convolution neural network recognition was significantly higher than that of the OCR software identification.The recognition character image provides the technical support for the image recognition system optimization.Image recognition can be used in the field of information storage,license plate recognition and verification code identification,which means that the study of image recognition will become more meaningful.