Font Size: a A A

Unstructured Text Detection And Recognition For Contest Certificates And Its Applications

Posted on:2023-12-20Degree:MasterType:Thesis
Country:ChinaCandidate:H M MaFull Text:PDF
GTID:2558307070982009Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
It is of great significance to accurately obtain the text information of the award certificate in the process of college student management.Certificate text extraction based on image processing is one of the main methods.Due to the problems of image degradation,format diversity and less marked data in certificate image,the localization of Chinese text information in image is not accurate,the confidence of text detection and the accuracy of text recognition are low.Therefore,this thesis mainly studies the unstructured text detection and recognition algorithm of contest certificate and develops the contest certificate management system.The main research work and innovative achievements are as follows:(1)Based on the preprocessing of contest certificate image contour search,filtering and noise reduction,a CTPN text detection algorithm integrating Anchor regression is proposed.CTPN backbone network is used to extract the text features of images,and the loss function of fusion Anchor regression is used to optimize the text features.The internal recursion mechanism is combined with the output feature graph to connect the real text box.Transfer learning strategy is introduced in the training process to solve the problem of insufficient certificate samples.Experimental results show that the proposed method has a confidence of 0.9655 for all samples.(2)Aiming at the low accuracy of text recognition and the difficulty of training convergence of BI-LSTM model,a bi-LSTM text recognition algorithm with Residual was proposed.The forward and backward information in sequence text is acquired by bidirectional long and shortterm memory network,residual module is introduced in input layer and output layer respectively,and attention mechanism is introduced in the decoding part of CTC framework to decode feature sequence,so as to realize text recognition.Experimental results show that the accuracy of the proposed method is 93.22%.(3)The competition certificate management system is designed and developed.Based on the demand analysis of the management system,the overall framework structure,back-end technical architecture and front-end technical architecture of the management system and the collection and application process of the award-winning information have been designed to achieve the collection of award-winning information,award-winning information statistics,comprehensive evaluation and application and other functions.The developed system can automatically extract the text information from the competition certificate and improve the daily management level of college students.
Keywords/Search Tags:Information system, Text detection, CTPN, Text recognition
PDF Full Text Request
Related items