Automatic Recognition Of Basic Formula Under Mobile Phone Photographing

At present,the Internet education industry is developing rapidly,and there is a need for automatic identification and automatic correction of mathematical expressions that appear in the Internet teaching process.To automatically identify and correct such basic formulas,it is necessary to locate,segment,and identify the formulas in the picture,and finally convert it into text data and input it into the computer system for automatic correction.However,because the basic formula recognition in the teaching stage has certain special characteristics,especially the general vertical type in the elementary school stage has a certain complexity in the positioning and segmentation process.The basic vertical formulas for detection and recognition has particularity and certain difficulty.It is not easy to realize positioning and recognition by recognizing conventional traditional text.Therefore,it is very meaningful to research and implement a basic algorithm recognition system based on the mobile phone camera.This thesis makes an in-depth study on the formula positioning and recognition technology involved in the basic formula recognition system,and uses the deep learning network model to locate the basic formula.Based on this as a research foundation,this thesis designs and implements a computer-based recognition and correction system for basic formulas under mobile phone photography,which provides a feasible and efficient computer-based online teaching correction program for elementary school basic formulas,helping teachers and students to learning online.The main contents of this article are as follows:(1)In this thesis,for the actual application scenarios of the basic formula recognition and correction in the elementary school stage,1500 pictures containing basic formulas taken by mobile phones are collected as data sets.These pictures contain a total of 10254 basic formulas,and the fonts,structure,and layout of the formulas are diverse.In addition,we label the basic formula data set,and the labeling information includes the formula positioning frame,formula identification content,and formula results.(2)In this thesis,based on the current deep learning theory,according to the relevant needs of basic expression recognition in the mobile phone camera scene,the CTPN text recognition model is targeted to improve,and based on this,a basic formula detection positioning model is proposed.This model is superior to other deep learning network models in detecting and locating text line regions of basic formulas.Aiming at the basic vertical formulas of elementary school,this thesis uses the clustering method to connect the vertical text lines in context,accurately identify the basic vertical formula process and make corrections.In this thesis,based on the research of basic formulas detection and recognition,the Java Spring framework and Python language are used to implement a automatic basic formula recognition system based on the basic formula detection and positioning model and the open source recognition engine Tesseract.The system has five main modules:the image preprocessing module,the formula positioning module,the vertical connection module,and the formula recognition module,and the result recombination module.The image preprocessing module is mainly responsible for image noise removal,size compression adjustment and rotation correction.The formula positioning module uses the improved CTPN model to detect and locate the text lines of the algorithm,and vertically connects the image areas with the basic vertical form to establish the context correspondence in the basic vertical recognition process.The formula recognition module uses the open source recognition engine Tesseract to identify and exclude the formula text box,and then identify and correct the miscalculated image.Finally,the result recombination module reorganizes the recognition result,and returns the result to the mobile terminal,and the mobile terminal displays the result to the user.