The Research And Application Of Deep Convolution Neural Network In OCR Problem

Optical character recognition(OCR)is now mainly used in document recognition and document identification.Document recognition can digitize printed documents to quickly and accurately extract valid information.Document identification digitizes scanned documents or copies of documents to improve work efficiency and reduce work intensity.As a branch of the field of artificial intelligence,deep learning can improve the scope of application of OCR recognition,and text area extraction applied to OCR can enhance the accuracy of OCR extraction of text and improve the accuracy of OCR.Analyzing and researching three key issues of OCR: text area detection,character cutting and recognition,and optimization of three problems combined with deep learning.Firstly,studying the possible problems of the three key points in practice,and base on the methods of convolutional neural network to make appropriate improvements and optimizations.Finally,the OCR recognition system based on this research method is completed.A text region extraction method based on deep convolutional neural network for document images is proposed and implemented.A method based on deep convolutional neural network is used to detect images using fixed-width,different-height text sequence boxes,and then combine text sequence boxes into text lines as output.In this process,the precise positioning and accurate positioning of the text region in the image is the highlights.On the other hand,the use of deep convolutional neural networks for text area detection is a huge challenge at speed.Based on the fast region convolutional neural network method,adopting some of the algorithm ideas to avoid a large number of repeated convolution calculations,taking into account the efficiency and accuracy,making it possible to achieve a real-time document identification system.Using the binarization method for document image text line image and presenting a character segmentation method based on neural network feedback,the character segmentation is optimized.This study found that the difficulty of character cutting mainly lies in the complex scene caused by the mixed arrangement of Chinese characters,English and numbers.Through the research on the character cutting method,a method based on neural network is used to determine the extracted text lines based on the pixel points and binarization.Then the vertical projection method is used to cut the characters to improve the accuracy of OCR recognition result.A character recognition model based on deep convolutional neural networks for document images is proposed and implemented,and a training data set for ID images is constructed.The difficulty in identifying this key point lies in the problem of the recognition accuracy caused by the large number of categories.This paper studies various methods of OCR recognition,and proposes a recognition model based on deep convolutional networks to improve its accuracy.And optimize with context in the end.An OCR document identification system is realized based on the above-mentioned theories of various technologies.The system is mainly divided into text area positioning,character segmentation and recognition.Through experimental analysis,the deep convolutional neural network method used in the regional positioning of this paper can have better generalization processing capabilities.