Master Thesis Defense: Parmida Atighehchian

Speaker: Parmida Atighehchian

Supervisor: Dr. C. Y. Suen

Examining Committee:
Drs. T. D. Bui, B. Jaumard,C. Poullis (Chair)

Title: Coin Wear Estimation and Automatic Coin Grading

Date: Monday, April 3, 2017

Time: 14:00

Place: EV 11.119

ABSTRACT

In numismatic studies, coin grading is referred to as the set of detailed experiments on a coin in order to estimate its quality, which is the most important factor to estimate the coin's value. Usually, the task is done by three expert numismatists to minimize personal biases. Each numismatist tests the coin's wear, coloration, and toning under different lighting conditions. Coin grading is a sensitive task to be done by humans. There are different parameters that can define the coin's value, however, dependent on the numismatist expert conducting the test, some parameters are neglected and some are given a heavier weight, which makes the procedure very subjective. A computer-aided algorithm for coin grading is considered an asset to help conduct more objective coin grading experiments.

We propose a coin wear estimation algorithm, which is fully based on features extracted from the digital images of coins. Apart from coin grading, the proposed algorithm is useful to find and dismiss the heavily worn out currency from the market. As online trading is getting more and more popular among coin collectors, it has become easier for individuals to sell a low-quality coin instead of a high-quality one or foist fake copies instead of real coins. This study is concentrated on the feasibility of having a computer-aided program to conduct coin grading. The required specifications for the dataset are fully investigated and the final dataset is collected after lots of experiments. In our proposed method, SIFT key points are used to distinguish the amount of wear on the coins. These key points are known for their high accuracy in shape detection problems. Our approach in using these descriptors to estimate the amount of wear on the coins attains a high accuracy of 93%.