Student Handbook Recognition System Using Deep Learning

Digital education information management systems have been utilized by a variety of schools and educational enterprises for better managing student and teacher information with the development of education informatization.However,it can be found that student handbooks for primary school students are primarily stored in paper forms.It will spend a large amount of time and manpower to manually input those handbooks into the system.In order to reduce time and manpower costs,a student handbook recognition system based on deep learning has been designed and implemented in this paper,so that automatic entry of student handbook information can be achieved.The proposed student handbook recognition system is designed and implemented against a specific student handbook.The common table location algorithm and character recognition algorithm have high requirements on image quality and writing specification.However,the table format of different pages of the student handbook in this paper is different,and due to the influence of the process of using and collecting,the table of the student handbook page is bent and tilted,which is not conducive to the table location.In addition,students and teachers have different ways of writing,and there exist unclear writing and touching characters,which are not conducive to character recognition.In view of the above situation,this paper designs and implements a student handbook recognition system,which consists of three parts,namely image preprocessing,table cell location and character recognition.The following three aspects are main novel contributions of this paper.(1)Image preprocessing and table cell location algorithms are studied.Complicated structure,diverse forms,and varied filling methods in the student handbook image lead to multiple imaging problems,which cannot be handled with a single image processing method.In this paper,a targeted image processing flow is designed in this paper on the basis of studying functions and principles of algorithms related to image preprocessing and table cell location.Consequently,the IOU of the subsequent table cell position task can be raised by 2.66%.(2)A table cell location algorithm based on deep learning is proposed.Based on deep learning,the table cell location algorithm is proposed against low IOU,poor robustness,poor expansibility and the like problems that can be found in the interference of the student handbook image when the traditional table cell location algorithm is adopted.Its IOU is 96.32%,being raised by 5.20%.(3)An improved character segmenting algorithm is put forward.A secondary extraction algorithm is proposed in this paper to cope with the problem of segmenting small-sized page numbers from large-sized student handbook images.Specifically,rough extraction is firstly performed in line with the distribution of page number areas.Next,page number areas are extracted finely with sliding windows,implementing the segmentation of page number characters.In addition,a projection segmenting algorithm of self-adaption thresholding is also put forward to handle the problems of table lines interference and touching characters interference in segmenting the character of a single table cell,which eliminates the table line and touching characters areas based on the projection result of the table cell image.In this way,table cell character can be segmented.Finally,the accuracy of page number recognition,rank recognition and evaluation recognition were 96.12%,95.42% and 96.75% respectively.39 figures,3tables,and 41 reference articles are contained in the dissertation.