Author: Guo, Qiming
Title: Research and simulation of PCA and SPP algorithms to improve CNN character recognition performance
Advisors: You, Jia (COMP)
Degree: M.Sc.
Year: 2020
Subject: Pattern recognition systems
Neural networks (Computer science)
Optical character recognition devices
Hong Kong Polytechnic University -- Dissertations
Department: Department of Computing
Pages: v, ii, 66 pages : color illustrations
Language: English
Abstract: Handwritten character recognition is one of the researches and application fields of computer vision based on Deep Learning. In the future, Handwritten character recognition will continue to be used in education, industry, finance, banking, home use, smart devices, and other fields. With the advent of the era of big data, the demand for handwritten character recognition has increased dramatically. Because handwritten characters themselves provide less information, and the font styles of different writers vary greatly, some handwritten characters are still difficult to be recognized by humans, so handwritten character recognition has always been one of the research hot spots. With the extensive use of mobile devices, more character recognition scenarios are using mobile phones and other mobile terminal cameras to recognize non-standard characters. The characters to be recognized have different sizes and different viewing angles, and there may be deflection and distortion. In this regard, although there are continual updating and iterative convolutional neural networks, such as FAST-RCNN, it cannot be solved very well. Studying how to improve the character recognition rate and recognition speed are still two key issues to be considered. In order to study this specific recognition task, this paper proposes a method of introducing PCA-Net and SPP-Net based on CNN. It is hoped that the introduction of PCA-Net can improve the efficiency and recognition rate of character recognition, while SPP-Net can solve the problem of incompatible size and the problem of increased error rate in character. The research results show that the method combining PCA and CNN has fewer iterations and high recognition rate when the same mean square error occurs; for single-scale training, the method combining SPP and CNN has high recognition rate (about 2.2%); Training combined with SPP and CNN methods can recognize images of different scales at the same recognition rate. The research results not only verify the effectiveness and feasibility of the two methods, but also lay a certain foundation for further research.
Rights: All rights reserved
Access: restricted access

Files in This Item:
File Description SizeFormat 
5806.pdfFor All Users (off-campus access for PolyU Staff & Students only)2.28 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: