Machine Vision/Computer Vision Card App

[Mark17]

Originally Posted by drcy

Focus on authentication, skip grading.

Totally agree. It's the grading aspect that's at the root of the entire problem. Determining authentic/counterfeit and original/altered is all you should focus on. Once the grading aspect is included, the nefarious games will begin again (people looking for a way to obtain higher grades by fooling your machine.)

[kevinlenane]

Honestly authentication is likely much much easier anyway - it's always easier to start w a narrower problem and i am certain I can nail Goudeys and a ton of other pre-war given their characteristics. Generally I think it could be done w some kind of standardized image capture - potentially from a phone for mass usage but I'll probably start w a high res scan format as an input. Cut lines show up pretty well in 2d but a profile image could work too. In general you want an image input thats easy for everyone to produce..

I'll See what I can get going in the Jan/Feb and I'll post it here when it's ready though I will for sure need training images at some point...

[Bigdaddy]

Originally Posted by Mark17

Totally agree. It's the grading aspect that's at the root of the entire problem. Determining authentic/counterfeit and original/altered is all you should focus on. Once the grading aspect is included, the nefarious games will begin again (people looking for a way to obtain higher grades by fooling your machine.)

Agree, focus only on verifying authenticity of the card and whether or not it has been altered. Alteration can be a bit more nuanced though as it could include things like trimming, soaking, erasures, wax/stain removal, recoloring, addition/subrtaction of material, etc.

I would leave the grading, a totally subjective measure, to the eye of the beholder. Sort of like beauty.

[kevinlenane]

I'll post image specs when I get to the training phase. I think the interesting thing will be to get some images of good fakes beyond the Gelasso and Inkjet stuff. I have a couple of auth markers which should be pretty hard to fake. I also wonder if the methodology should be posted or not but in the initial run I'm going to add code comments that describe exactly what is going on for testing purposes. Stay tuned....

[Bigdaddy]

Have you thought about copyright or patents?

There is a lot of open source software out there these days, but it might be best to make only the code available in executable form rather than source code.

In general, you don't want the bad guys to know where all your defenses are.

[BabyRuth]

Baseball Card Collecting - Rise of the Machines

[steve B]

This is pretty interesting.

I was talking about this (again, she must be getting sick of it) with my wife who is in the computer industry, and I was explaining how the sheer number of sets with varying degrees of challenge in picking out real vs fake automatically made it a difficult problem.

Her take was that it would be easier than I think, if it was done with a decent image library, some machine learning, and at first a person who knew the set essentially teaching the machine how to tell.

This seems to be essentially the same thing, perhaps without the human teacher, but working on one set that isn't too big, and that's also pretty consistent, with few if any small differences between examples of the same card. Unlike say T206, where each front comes in at probably at least 6 distinctly different versions.

I've been down on previous ideas for this, but I think this one could work.