Research On Text Detection And Recognition Algorithms For Mathematical Exercise Book Image Based On Deep Learning

Information extraction from text images has become a key part of intellectualization with the development of technology.As Optical Character Recognition(OCR)has also been proved as an indispensable part in various industries,technology nowadays is turning into intellectual detection and recognition from traditional manual excerpt and recognition,and has achieved good results on multiple experiments.It is also widely used in industrial areas such as text recognition,text on photo detection,receipt recognition and number plate's recognition.In this paper,we have successfully developed an auto-grading system using text positioning and recognition on photos which can be used in homework grading use cases.This paper also analyzed OCR theories,exiting deep learning in image detection and recognition,and applied some recent OCR detection algorithms such as MSER,CTPN and OCR recognition algorithms such as CNN,CRNN.Meanwhile,the paper performed image correction,algorithm improvement,feasibility testing,system building,operating practice,and obviously shortened time period of deep learning in training.By changing the RNN structure of the detection and recognition model to the Conv1 D structure,and theoretically giving an explanation.The new framework,by optimization on detection and recognition modules,achieved 4 times quicker than the original training speed of CTPN in detection and 6 times the training speed of CRNN in recognition on product dataset.And the accuracy of the two processes did not change much.