I've been working on a small project for the excellent girlfriend involving the recognition of coins. I've been looking into some various different applications that attempt to do this using computer vision. Turns out its a hard program.
I've also learnt a new word, clearly!
A small program called Scybot attempts to count US coins with the iPhone camera. Sadly, it doesnt work. It turns out it simply uses a Hough Transform by the looks of things. I tried this and here are the results:
Sadly, it isn't great on coins that aren't round or at least, not easily. It seems that perhaps training a haar classifier might be the best option? This blog post is a great introduction to the idea of training a classifier. It's an intensive task and the jury is still out on whether or not this will actually work. I'll be sure to post my results here when I get them.
Currently, my classifier is still running, as has been for the last 3 days!
At the moment, most of the research seems to be based around the recognition of old coins as oppose to new ones?