Outed cards, now including a 130K gain on an Aaron rookie

[steve B]

Originally Posted by Bigdaddy

Ummm.....facial recognition is a current technology that can tell who you are from a picture. It is used on a daily basis by many organizations. Put on glasses, grow a mustache, get a suntan..doesn't matter, facial recognition will see through the distractions to identify a person. And that is not cutting edge technology, it is the technology of today, in use around the world. An artificial intelligence system based on machine learning could do this job much better and much more repeatably than a human ever could. Have you seen the recent commercial where computers pick out snow leopards from images taken by trail cameras? And the comparison of how long it would take a human to go through the same amount of photos? Current image processing technology is way beyond what a human can do with their eyes. Especially for things like noticing flaws in flat two-dimensional objects, detecting freshly trimmed edges or undersized cards and doctoring.

It can only identify someone in its database. Currently, the only photo databases for cards are scattered among several different websites, and as far as I know, none are detailed enough to allow a determination between real and fake.

Identifying what the card supposedly is should be easy under some conditions. Card in a database, card in good enough condition. Card that doesn't have a lot of variance in its manufacture.

Here's an couple example of ones that would be a bit harder 48 Leaf (Yes Ted, I know they're really 49.. ) Here's a screenshot of my spreadsheet I did showing most - or as many as I could find - of the different varieties.



Just the Jensen has four distinctly different ones, different hat shape, different background, probably some other difference I haven't spotted, or haven't got an image of yet (Many come with shaded hats, but I haven't seen a shaded hat Jensen.

That database will need all of those to even have a chance of being right about the real/fake question. And it would probably be wrong about a shaded hat version if it came across one, because none of the "real" ones has shading there.

Most of the junkwax era stuff especially Fleer have cropping varieties for nearly every card. Both front and back. They are almost entirely uncataloged, so while you may be able to tell that card is a particular card, if it's not cropped the same as the one in the database... your system should reject it as fake.

88 Score... Has three different die cuts. The first left small tufts of fibers on each corner. When they aren't severe, they can look like wear. The second moved the tufts out about a quarter inch from the corner. The third, they finally got the die cutting right. Oh, and they screened the different printings differently.... None of that is in any catalog.

The Magie T206 comes in at least 8 distinctly different varieties. (All T206s will, some are just much harder to tell.) Not cataloged, there's maybe a handful of people who have even cared to check.


And,

Cards are not 2 dimensional. One of the ways to tell if certain cards are trimmed is that the edge doesn't have a rounded side and a ridged side.
And slight paper loss can be very hard to pick up in a scan, and sometimes even when the card is in hand. (see the crackerjack variation thread, is it paperloss, or printed without a number, or some other sort of misprint? Turns out it's a bit of paperloss that over a long time has become smooth enough to pass. Or maybe a misprint... Once it's determined to not be a new variation, the exact reason really isn't looked at all that closely.

For most cards, we don't even have a good reference for how thick they should be. And thickness is difficult to measure from a scan.

In comparison, the "snow leopard or a rock" question is easy. Especially if you have a good set of photos of the rocks you're looking at.

The "is that a terrorist or a tourist" is a bit tougher, and as Pete pointed out, a non-government portable commercial product that only has to identify ONE face isn't entirely reliable.