Login | Register

Selection of Robust Features for Coin Recognition and Counterfeit Coin Detection

Title:

Selection of Robust Features for Coin Recognition and Counterfeit Coin Detection

Al-Frajat, Ali Khadair Kadam (2018) Selection of Robust Features for Coin Recognition and Counterfeit Coin Detection. PhD thesis, Concordia University.

[img]
Text (application/pdf)
Al-Frajat_PhD_S2019.pdf - Accepted Version
Available under License Spectrum Terms of Access.
5MB

Abstract

Tremendous numbers of coins have been used in our daily life since ancient times. Aside from being a medium of goods and services, coins are items most collected worldwide. Simultaneously to the increasing number of coins in use, the number of counterfeit coins released into circulation is on the rise. Some countries have started to take different security measures to detect and eliminate counterfeit coins. However, the current measures are very expensive and ineffective such as the case in UK which recently decided to replace the whole coin design and release a new coin incorporating a set of security features. The demands of a cost effective and robust computer-aided system to classify and authenticate those coins have increased as a result.
In this thesis, the design and implementation of coin recognition and counterfeit coin detection methods are proposed. This involves studying different coin stamp features and analyzing the sets of features that can uniquely and precisely differentiate coins of different countries and reject counterfeit coins. In addition, a new character segmentation method crafted for characters from coin images is proposed in this thesis. The proposed method for character segmentation is independent of the language of those characters. The experiments were performed on different coins with various characters and languages. The results show the effectiveness of the method to extract characters from different coins. The proposed method is the first to address character segmentation from coins. Coin recognition has been investigated in several research studies and different features have been selected for that purpose. This thesis proposes a new coin recognition method that focuses on small parts of the coin (characters) instead of extracting features from the whole coin image as proposed by other researchers. The method is evaluated on coins from different countries having different complexities, sizes, and qualities. The experimental results show that the proposed method compares favorably with other methods, and requires lower computational costs.
Counterfeit coin detection is more challenging than coin recognition where the differences between genuine and counterfeit coins are much smaller. The high quality forged coins are very similar to genuine coins, yet the coin stamp features are never identical. This thesis discusses two counterfeit coin detection methods based on different features. The first method consists of an ensemble of three classifiers, where a fine-tuned convolutional neural network is used to extract features from coins to train two classifiers. The third classifier is trained on features extracted from textual area of the coin.
On the other hand, sets of edge-based measures are used in the second method. Those measures are used to track differences in coin stamp’s edges between the test coin and a set of reference coins. A binary classifier is then trained based on the results of those measures. Finally, a series of experimental evaluation and tests have been performed to evaluate the effectiveness of these proposed methods, and they show that promising results have been achieved.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Computer Science and Software Engineering
Item Type:Thesis (PhD)
Authors:Al-Frajat, Ali Khadair Kadam
Institution:Concordia University
Degree Name:Ph. D.
Program:Computer Science
Date:September 2018
Thesis Supervisor(s):Suen, Ching
ID Code:984864
Deposited By: Ali Khadair Kadam Al-Frajat
Deposited On:10 Jun 2019 13:45
Last Modified:10 Jun 2019 13:45
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Back to top Back to top