Quote:
Originally Posted by Bigdaddy
As I've said before, this is a grad student project. Maybe a team, but surely something that some students at one of our leading universities could tackle in short order. There are two keys, the first of which is to come up with a set of rules to grade the cards. Corner wear, wrinkles, creases, surface wear, centernedness (?), etc. And how do they add up to a grade. People keep saying that the human is required for 'subjectiveness', but isn't that what we want to eliminate? Consistency and a set of codified rules is what I want.
The second is a method to detect alterations - UV light, magnification, edge examination, ink patterns, etc. A couple of different sensors could accomplish this and then train the machine with a good set of unaltered cards (machine learning) and a set of known altered cards so that it can detect alterations. What are the min/max measurements of a factory cut. What should the edges look like? They should be at least as dirty/worn/frayed as the surface.
Both of these techniques could be automated and it would take the human part out of grading. No need to resubmit, looking for a bump. And it would take no more effort to grade a T206 than a 1978 Topps. Cost to grade would not be a function of the cost of the card.
if a company could do this, they could set up a registry for their own cards and include PSA (and SGC, Beckett, etc.) also, but levy a -2 pt bump on any card not graded by them.
This is not rocket science folks. We have carbon dating, we can find criminals through their relative's DNA samples, Facebook knows who is in your pictures before you tag them, cars can drive themselves and you can carry on a conversation with a $35 computer (Alexa). To have people still grading cards is like having a corded, rotary phone in your house. We can do better.
|
There are a lot more variables than you would think when it comes to alterations. The data on actual card size hasn't really even been gathered yet.
What's correct for a T206 isn't for a 1978 Topps.
And when you get to more modern cards, the edge quality can be different on different edges of the same card.
I can think of at least 4-5 sets that would be difficult, even for a computer, assuming it could do the imaging.
Consider a simple set. Something with no varieties, and few cards.
Maybe 78 Zest soap. Only 5 cards. And not expensive. Getting a nice high res scan for comparison wouldn't be hard, so far so good.
But for a device that compares things to see of they're exactly alike, if one has the magenta printed more towards the top than the blue, and the one the machine is grading doesn't, they will appear to be different. So you'll need some leeway for slight misregistration. say the scan is 1200dpi, that means you can measure the misregistration down to 1/1200th of an inch. Way finer than the factory guys could or would. How much misregistration is too much? And that needs to be figured for all four colors.
A slightly more complicated set?
Most of the early UD sets have at least two different holograms. And most scanners have a very difficult time imaging them. And scratches can make them sort of illegible.
More complicated?
88 Score has three different die cut patterns, and each color group has at least two different angles of screening.
T206 is not four colors, not even the usually claimed six, but more like eight. And the registration is usually off. And plate wear and inking levels and changes in the actual art add a bit more.
1993 UD has - for a portion of the set- three different gloss patterns. And they are really hard to image.
Most of the modern Gypsy queen sets have cards that are knife cut on at least one side, but die cut on the others. Another imaging challenge.
So there's 6 sets in a variety of complexity.
By the time the kids can code that and figure out how to do the imaging, I could probably authenticate thousands... And can teach someone else to do the same.
Good luck, currently I'll take people over the machine.
(Recently had a college get a grant to work on a project. 3 months, 30K. They accomplished... learning that one part was proprietary, and that the company wouldn't license it. And not much more. )