Threshold center-symmetric local binary convolutional neural networks for bilingual handwritten digit recognition

Saleh Al-Wajih, Ebrahim Qasem (2022) Threshold center-symmetric local binary convolutional neural networks for bilingual handwritten digit recognition. Doctoral thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p EBRAHIM QASEM SALEH AL-WAJIH.pdf Download (841kB) | Preview
	Text (Copyright Declaration) EBRAHIM QASEM SALEH AL-WAJIH COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (525kB) | Request a copy
	Text (Full Text) EBRAHIM QASEM SALEH AL-WAJIH WATERMARK.pdf Restricted to Registered users only Download (33MB) | Request a copy

Abstract

Arabic and English handwritten digit recognition is a challenging problem because the writing style differs from one writer to another. In middle east countries, many official forms are prepared to be written using either Arabic or English languages. However, some people fill the form using both languages (Arabic and English), which adds more challenges to recognize digits. Nowadays, deep learning approaches are considered the hot trend of new research, including Convolutional Neural Networks (CNN). CNN is used in many applications and modified to produce other models such as Local Binary Convolutional Neural Networks (LBCNN). LBCNN was created by fusing Local Binary Pattern (LBP) with CNN by reformulating LBP as a convolution layer called Local Binary Convolution (LBC). However, LBCNN suffers from the random assign 1, 0, or -1 to LBC weights, making LBCNN less robust. Nevertheless, using another LBP-based technique such as Center-Symmetric Local Binary Patterns (CS-LBP) can address such issues. In this thesis, a new model based on CS-LBP is proposed called Center-Symmetric Local Binary Convolutional Neural Networks (CS-LBCNN) that addresses the issues of LBCNN. Further, an enhanced version of CS-LBCNN is proposed called Threshold Center-Symmetric Local Binary Convolutional Neural Networks (TCSLBCNN) that addresses another issue related to the zero-thresholding function. The proposed models are compared against state-of-the-art techniques that used the MNIST and MADBase as a bilingual dataset. The proposed TCS-LBCNN model proves its ability to give a more accurate and significant classification rate than the existing LBCNN models. For the bilingual dataset, the TCS-LBCNN enhances the performance of LBCNN and CS-LBCNN, in terms of accuracy, by 0.15% and 0.03%, respectively. In addition, the comparison shows that the accuracy acquired by TCS-LBCNN is the second-highest using the MNIST and MADBase datasets.

Item Type:	Thesis (Doctoral)
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Divisions:	Faculty of Computer Science and Information Technology > Department of Information Security
Depositing User:	Mrs. Sabarina Che Mat
Date Deposited:	26 Feb 2023 07:09
Last Modified:	26 Feb 2023 07:09
URI:	http://eprints.uthm.edu.my/id/eprint/8412

Actions (login required)

View Item