Character Recognition From Instructional Video Based On Convolution Neural Network

With the enormous development of information technology and the advent of the era of Internet Plus.In education sector,an Internet-based technology evolution is happening quietly,that is the emergence of the massive open one-line course(MOOC).In recent years,with more and more institutions of higher education joined MOOC and opened their MOOC courses,a growing number of students are taking courses via MOOC.Howerver,since most MOOC couses are video-based,students usually have to watch these videos repeatedly while learning courses,and it is a waste of time checking every frame of the video to search the contents they want.Therefore,it is of great siginificance to devise a method that can locate the enquiry content by retriving the content of each frame in course videos.This paper concentrates on one aspect of above methos,that is character recognition in course videos.Recognise Chinese characters in videos by convolution neural network,including:(1)For character recognition in course videos,due to its characteristics,the change of contents in each frame of videos is slow,and duplicated recognition can happen in character recognition,since we can make recognition from one of the houdrends of consecutive frames in the videos.So,how to get such frames is a problem.Given this condition,by considering len dectection,this paper propose a method that can position character by imposing image processing methods on key frames,which are got from consecutive frames by applying lence boundary based methods.(2)For the convolution neural network training data set,a Chinese data set is constructed by hand,which includes the print data of the different fonts generated by the program and the handwritten data from the CASIA-HWDB database.It is a random increase in noise points,distortion,blur and watermarking and other data sets to enhance processing,which is not enough for the change of the printed data.(3)For the training of convolution neural network,the structural design,activation function,loss function,learning rate and parameter initialization of the network are studied respectively.Finally,a robust convolution network model is trained,Used to recongnise Chinese characters(including handwriting and print).