Recognizing Rare Coins

My dad’s mother was a coin collector.  When I was seven or eight years old, she bought me a Whitman folder to fill with pennies from 1941.  Grandma Mischel taught me to always check my change for pennies that I needed to fill my book, and also for certain other coins (for example, dimes, quarters, and halfs minted before 1964, because they contain 90% silver).

For those difficult-to-find coins needed to fill my penny book, she suggested that I go to the bank and buy several rolls of pennies.  That worked, although I probably went through 20 rolls before I got all the coins I needed.  Filling the non-labeled spots in the book (probably four years’ worth–up through 1975) was trivial.  I just had to wait a few years.

When she was younger my grandmother would go through her change periodically and pick out the interesting ones.  She apparently stopped doing that at some time.  When she passed away, she left many jars full of coins.  As I recall, a sampling of the jars showed that most of the coins were worth no more than their face value and the jars were sold by weight.  Some collector ended up with thousands of coins that he had to sort through, by hand, to find the few that had any numismatic value.

Looking at my own change jar (Debra and I throw all of our change into a vase and then cash it in once it’s full), I got to wondering about whether there are any valuable coins in there.  I know there are a few coins that would carry a small premium (for example, a one-cent piece minted before 1959–a “wheat penny”–is typically worth two to five cents), but it’s unlikely that there are any coins that would carry a huge premium.

I could go through the coins manually and extract those few that are worth more than face value, but it’s a terribly tedious task, and I’d probably miss a few either through inattention or because I’m not a collector:  I don’t know enough to know which coins I should be interested in.  For example, until yesterday I’d never heard of the Wisconsin quarter error or the 1995 double-die Lincoln cent. 

Coin sorters have been around for decades, but most are purely mechanical devices that sort based on size.  I got to wondering if it’d be possible to create a coin recognizer that could look at a dime, say (identified as a dime by a coin sorter, most likely), and determine if it’s a Mercury dime or a Roosevelt dime, and read the mint date.  Based on the type and date, the machine could put the possibly rare coins in a different slot.  All of the common coins would get bagged or rolled.

There’s been some research on image-based recognition of ancient coins in an attempt to prevent or at least diminish the trade in stolen coins, and I found another paper called Classification of coins using an eigenspace approach, but I haven’t found anything commercially available.

How hard would it to be to build such a thing?  How quickly could it classify coins?  It’d be a very interesting project to work on, and it’s kind of fun to contemplate the possibilities, but would this device be practical?