Analysis of methods for the recognition of Indian coins: A challenging application of machine vision to automated inspection

The subject of this paper is a particularly challenging machine vision (MV) based sorting application where the `part' is an Indian coin. The application is challenging in part because of the lack of distinctive features to differentiate between denominations as well as the variability in the features for a given denomination. Although there are coin recognition algorithms documented in the literature, the applications are typically tested off-line with static images of the coins. In this paper, a MV-based system for on-line recognition and counting of Indian coins moving on a conveyor is evaluated. The accuracy and performance of three different techniques are compared: particle classification, pattern matching and geometric matching. The conclusion is that none of these three techniques produced acceptable results, where the goal was to achieve 95% accuracy at 1000 coins/min.