An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique
Scanned Document Authentication (SDA) is one of the major areas in authenticating document copies, such as tickets, passport pages, or certificates based on the document fingerprints. These fingerprints are represented by the textual features extracted from the scanned documents. Existing SDA method...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English English English |
Published: |
2022
|
Subjects: | |
Online Access: | http://eprints.uthm.edu.my/8485/1/24p%20SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI.pdf http://eprints.uthm.edu.my/8485/2/SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/8485/3/SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI%20WATERMARK.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
id |
my-uthm-ep.8485 |
---|---|
record_format |
uketd_dc |
spelling |
my-uthm-ep.84852023-04-02T00:52:58Z An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique 2022-07 Khaleefah Al-Dulaimi, Shihab Hamad T61-173 Technical education. Technical schools Scanned Document Authentication (SDA) is one of the major areas in authenticating document copies, such as tickets, passport pages, or certificates based on the document fingerprints. These fingerprints are represented by the textual features extracted from the scanned documents. Existing SDA methods mainly use local feature descriptors like local binary patterns and global feature descriptors like Gabor filters. However, the selection of these descriptors and their parameters’ configuration is dependent on the document type and the deformations due to the effects of printing, scanning, usage, and time. To address the dynamic nature of the document deformation, this thesis proposes a new Automated Scanned Document Authentication (ASDA) model that operate based on Collaborative Reinforcement Learning Agent (CRLA) architecture. The proposed ASDA model consists of four feature descriptor agents running on three phases, which are perception, decision, and action to control the operational behaviour of the feature extraction operators. ASDA performs feature matching using the Euclidean Distance and adjusts the feature descriptor agents and their parameters’ configuration according to the matching results. Meanwhile, the CRLA architecture implements a Q-learning strategy to direct the dynamic selection decision of the agents toward the optimum combinations of descriptors and configurations. The selection decision either leads to rewarded or penalty that guide the learning process of the agents with the goal to maximize the rewards results. The performance of the proposed ASDA model is evaluated using the UTHM-EDD and UKM-SPF datasets, which contains shear, lighting (reflection and transmission), crumpling, and printing deformation in three different resolutions of 50, 100, and 150 dpi. The ASDA model achieved an average accuracy of 98.40% for varying deformation conditions in the UTHM-EDD dataset and 100% in the UKM-SPF dataset as compared to other models; ELM, KNN, and OSELM. The results concluded that ASDA is capable to authenticate digital documents under varying degree of deformations with high accuracy. 2022-07 Thesis http://eprints.uthm.edu.my/8485/ http://eprints.uthm.edu.my/8485/1/24p%20SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI.pdf text en public http://eprints.uthm.edu.my/8485/2/SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI%20COPYRIGHT%20DECLARATION.pdf text en staffonly http://eprints.uthm.edu.my/8485/3/SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI%20WATERMARK.pdf text en validuser phd doctoral Universiti Tun Hussein Onn Malaysia Fakulti Sains Komputer dan Teknologi Maklumat |
institution |
Universiti Tun Hussein Onn Malaysia |
collection |
UTHM Institutional Repository |
language |
English English English |
topic |
T61-173 Technical education Technical schools |
spellingShingle |
T61-173 Technical education Technical schools Khaleefah Al-Dulaimi, Shihab Hamad An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
description |
Scanned Document Authentication (SDA) is one of the major areas in authenticating document copies, such as tickets, passport pages, or certificates based on the document fingerprints. These fingerprints are represented by the textual features extracted from the scanned documents. Existing SDA methods mainly use local feature descriptors like local binary patterns and global feature descriptors like Gabor filters. However, the selection of these descriptors and their parameters’ configuration is dependent on the document type and the deformations due to the effects of printing, scanning, usage, and time. To address the dynamic nature of the document deformation, this thesis proposes a new Automated Scanned Document Authentication (ASDA) model that operate based on Collaborative Reinforcement Learning Agent (CRLA) architecture. The proposed ASDA model consists of four feature descriptor agents running on three phases, which are perception, decision, and action to control the operational behaviour of the feature extraction operators. ASDA performs feature matching using the Euclidean Distance and adjusts the feature descriptor agents and their parameters’ configuration according to the matching results. Meanwhile, the CRLA architecture implements a Q-learning strategy to direct the dynamic selection decision of the agents toward the optimum combinations of descriptors and configurations. The selection decision either leads to rewarded or penalty that guide the learning process of the agents with the goal to maximize the rewards results. The performance of the proposed ASDA model is evaluated using the UTHM-EDD and UKM-SPF datasets, which contains shear, lighting (reflection and transmission), crumpling, and printing deformation in three different resolutions of 50, 100, and 150 dpi. The ASDA model achieved an average accuracy of 98.40% for varying deformation conditions in the UTHM-EDD dataset and 100% in the UKM-SPF dataset as compared to other models; ELM, KNN, and OSELM. The results concluded that ASDA is capable to authenticate digital documents under varying degree of deformations with high accuracy. |
format |
Thesis |
qualification_name |
Doctor of Philosophy (PhD.) |
qualification_level |
Doctorate |
author |
Khaleefah Al-Dulaimi, Shihab Hamad |
author_facet |
Khaleefah Al-Dulaimi, Shihab Hamad |
author_sort |
Khaleefah Al-Dulaimi, Shihab Hamad |
title |
An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
title_short |
An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
title_full |
An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
title_fullStr |
An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
title_full_unstemmed |
An improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
title_sort |
improved multi-agent system for scanned document authentication using collaborative reinforcement learning technique |
granting_institution |
Universiti Tun Hussein Onn Malaysia |
granting_department |
Fakulti Sains Komputer dan Teknologi Maklumat |
publishDate |
2022 |
url |
http://eprints.uthm.edu.my/8485/1/24p%20SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI.pdf http://eprints.uthm.edu.my/8485/2/SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/8485/3/SHIHAB%20HAMAD%20KHALEEFAH%20AL-DULAIMI%20WATERMARK.pdf |
_version_ |
1776103357075685376 |