Sorting And Variety Identification

In my opinion, grading is way too subjective (and complicated) for this type of automation, at least with today's technology. However, sorting would be a simple task as it involves simple pattern matching that is no different than identifying a finger print.

It does Numistudy by Jon Lusk.

The TPGs all grade coins by human eye - no automation is used. Yes, they tried some experimental systems in the 1990s, and filed a few patents. But they never worked well enough for market acceptability.

Just like they've had experimental robot cars since the 1980s, but you still don't have an autopilot in your car yet. They're just not reliable enough to work consistently in the real world.

If I understand your question correctly, I believe PCGS employs this technology in their new Secure Plus Grading ...



http://www.pcgs.com/secureplus.html

But they are using it for a different purpose. I suspect this software could be "tweaked" to add variety recognition and sort capability. Will it be offered to the collecting community? My guess ... not in the near future.

Completely agreed about the scale of the project, carl the just. Massive is an understatement once you get into variety identification. Ideally you'd start with sorting by year only, say 1938-2002 Nickels. First the machine must know the difference between 19XX and 20XX. You need either an OCR that can accurately read the text or a database of images that the machine can compare against. Next you'll focus on decades: 200X, 193X, 194X, 195X, 196X, 197X, 198X, 199X. Then finally the years. After you've taught all years of a specific series you're good to go on your date sorting. Anything that can't be identified with some high percent of certainty (95% for example) goes to a reject pile. The reject pile will house many things of interest: junk worn out coins that aren't legible, foreign coins and pre-1938/post-2002 Nickels, and any variety coins where the year is modified. At this point you can either start building up the ability to identify the outline of the coin's design, the other words on the coin, the mint mark, or any combination of these things. The more you develop your series the closer you're getting to being able to filter out anything with any small variations like you've mentioned. Now you can start building if/then functions; for example if the coin's text is doubled you can send it via your sorting machine into a variation bucket. The deeper you dive the more involved you'll become with teaching.

That's the process I see in my head. If the tools are built to support a GUI that anyone can use for teaching anyone would be able to jump in. I know in a manufacturing environment companies have software specifically for this. The software lets you identify priority features and tolerances and then outputs a pass/fail without any programming or technical input.

I'm probably just rambling about my daydreams and I apologize if you've managed to read through this and want your time back. :P

I laughed until I cried.

The software process described here would be extremely simple in execution. Even an average coder could do it; you don't need to duplicate spacing, just recognize numbers and letters. The physical setup - the moving of the coins - would be prohibitive in terms of manhours unless you used automation that would jeopardize the coins involved. And if the mint mark is on the reverse, your time is doubled for each coin.

Too funny. But if we're going BIO I think I'd rather have trained wasps doing the heavy lifting. With their air movement and small scale a non-intrusive cleaning operation could be included in the system. And I'm sure there aren't any fire or OSHA safety concerns with wasps/sq ft. OR labor laws about low wages and poor working conditions.

The machinery side would be interesting. If your software can't realign the coins before computing you'd need hardware that could rotate the coins for alignment. If the coin is fed upside down you'd need a flipping mechanism. If the MM is on the reverse and year on the obverse you'll need to capture both sides for analysis. The system would have to be zero impact, zero friction, and high speed. Those three things don't go well together. And the biggest challenge in my mind is keeping the system realistic (size and price) for the average consumer. I know I wouldn't be allowed to have a coin conveyor system running through the living room.