COIN-O-MATIC: A fast system for reliable coin classification

Systems that perform fast and reliable classification of heterogeneous coin collections can be beneficial to charity organizations and financial institutions that collect unsorted coins. Existing coin classification systems cannot classify heterogeneous coin collections. We present a new coin classification system designed to perform reliable classification of heterogeneous coin collections. In this case, reliability means with a low number of incorrect classifications. COINO-MATIC uses a combination of coin photographs and sensor information in the coin classification. The system preprocesses the coin photographs, and classifies the coins using edge-based statistical features. The classification is verified using a mutual information measure of the coin image and an averaged coin image that corresponds to the classification. We measure the performance of the system on a test set supplied by the MUSCLE CIS benchmark. We show that our system classifies approximately 72% of the coins correctly, while misclassifying only 2% of the coins. Moreover, the presented system is computationally efficient.