| The examination paper result review exists in each stage of education,and the traditional review method mainly depends on human labor.There are the following problems in the manual review of examination paper results :(1)teachers are high-quality talents,if a lot of time is spent in this work,it is a waste of human resources.(2)people repeat a thing for a long time,not only low work efficiency,but also increase the possibility of making mistakes.In order to solve the above problems,this paper designs a paper score automatic check system.The system is divided into two parts: cell data extraction and score recognition.The traditional score recognition method first binarizes the image and then extracts the score,and then divides the score into a single number for recognition.However,due to the complexity of the score extraction algorithm and the presence of concatenation between numbers,it is difficult to determine the segmentation point of the adhesive fraction and prone to false segmentation.In order to solve the above problems,this paper uses the color difference between the score and other content,and uses the mathematical morphology algorithm to realize the complete extraction of cell data.In order to avoid the improper segmentation of the score in the cell,this paper proposes an improved algorithm for handwritten number string recognition without segmentation under multi-classifier.The algorithm designs two kinds of classifiers,one length classifier and two number classifiers.Using the length classifier to obtain the character length information,the character length information is fed into the corresponding length number classifier.Through the joint cooperation of the three classifiers,the score recognition is completed and the result review is realized.The traditional lenet-5 network is improved by using the self-built isolated digital data set and concatenated two-digit data set,and three classifier models are trained.Comparative experiments show that compared with the traditional segmentation algorithm,the recognition accuracy of the unsegmented handwritten numeric strings with multiple classifiers is improved to some extent when the strings of 2,3,4,5 and 6 bits are recognized.The accuracy of examination paper review in this system is 98.2%,which is improved by 2.2%,7.71%,4.95% and 6.4%,respectively,compared with several review systems that use segmentation method to identify scores,and has certain practical value.The paper has 49 figures,11 tables and 55 references. |
