Our campuses are closed but thesis defences are proceeding normally, albeit remotely. Refer to our COVID-19 FAQs for more information.
When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.
Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.
The numismatic industry has been dealing with the problem of counterfeit coins for decades. With the growth of technology in counterfeiting, untrained users cannot distinguish fake from genuine coins, especially for rare and precious coins. Using coin experts is also very expensive, and even with them, in some cases, detecting counterfeit coin is not guaranteed. Consequently, the demand for a computer-aided system that can detect counterfeit coins and be robust and reliable has increased. This thesis proposes a new fake coin detection method that focuses on image mining approaches instead of extracting only statistical features from the coin image, as suggested by other researchers.
In this research, a novel framework called PrFA is proposed for counterfeit coin detection that shows the effectiveness of image mining techniques. We develop a new image mining system on top of the fuzzy concept that helps us to discover the implicit information from the images in the way closer to the human’s viewpoint. The advantage of the fuzzy set concept is that it can deal with uncertain objects, and we take this advantage for the decision-making problem by implementing an associative classifier model.
Our proposed framework is developed in two modules, and the principle of least privilege of it is a compressed and white-box system that can be considered as a knowledge attainment tool. In the first module, a method to detect the region of interests (ROIs) is applied that focuses on blob detection. In the second module, image mining is applied to find image patterns present in coin images using fuzzy association rules mining.
Image data are generally high dimensional due to a wide range of resolution levels. According to state of the art, the rule-based association methods demonstrated their efficiency by generating defensible solutions at an acceptable level of accuracy when dealing with small and medium-size samples. Regrettably, to cope with a large amount of data such as the image database, these methods were not robust enough. To tackle the above challenge, we propose a new algorithm for feature selection to reduce the dimensions of features via analyzing the relationships among different features.
Image classification is imperative to search for more available and appropriate information. In recent years, various methods based on image mining approaches for classification tasks have been explored. Apart from their usefulness, the available classifiers are often vulnerable to low accuracy. Accordingly, we present a pruned based fuzzy associative classifier algorithm to create a robust counterfeit coin detector system. This new classifier is a mixture of the association rules method and the fuzzy set concept.
In this study, we preserve the full power of fuzzy association rule mining to reduce the amount of redundant and insignificant rules by focusing on pruning methods. By comparing the achieved results with some other methods obtained from the same dataset, we demonstrate that our framework surpasses in terms of lower feature dimensions, and smoother boundaries while maintaining satisfactory accuracy. Besides that, we show that our proposed classifier is more accurate compared to other associative classifiers.
In this research, the problem with a general form will be described to provide a common framework for issues appearing in other domains.