Is it possible to create a laser-scan card grading machine?

[hilgum]

Wow - so much faith in quickly trained 20-somethings, and so little faith in established technology.

The problem with computerized card scanning and grading is not a technical one—it’s a business-case one. Computer-aided visual inspection has been a part of quality control in manufacturing for decades. You’d be hard pressed to find a sophisticated manufacturing enterprise that doesn’t utilize it. And meanwhile, the folks at Google and elsewhere are developing driverless-car technology based in part, and heavily reliant on underlying edge-detection and pattern-matching algorithms. I can use Digital ICE on my flatbed scanner to remove dust and scratches detected in scanned images, and then do further image editing in Photoshop, relying on its sophisticated image-analysis tools (think Magic Wand!). My iPhone camera focuses automatically. And a lot of this is virtually instantaneous. I would argue that the only real technical problem for card grading is that you would risk ending up with a system that would find—and make grading decisions on—details that are not actually visible to the naked eye.

Look at the grading standards for the major TPGs: centering, focus, sharpness of corners, breaks in surface gloss, stains, print or refactor lines. The numerical grades simply quantify the measurements of these features. If these features can be seen—and they can!—they can be quantified and factored into an evaluation function that can be tailored to any specific subgroup of cards you want to define (machine-cut, perforated, hand-cut, … T206, ’52 Topps, ’71 Topps, etc.).

But what’s the business case for such a system tailored to sportscard grading? Like everything else, it’s expensive technology, even when a lot of the hard technological work has already been done. As Pete (ullmandds) has mentioned, most cards have already been graded. This is an interesting observation, and I think largely true (ignoring all the cards now and forever more being produced, obviously). What’s the motivation for people to submit already graded cards to a computerized TPG? Andy (bn2cardz) is also right - there isn’t any “hope” for an anomalous upgrade from a deterministic algorithmic process. Why risk obtaining a lower rating on an existing PSA- or SGC-graded card. Why spend money to have a mid-grade card boosted, when it’s generally the case that a mid-grade card is a mid-grade card for an obvious reason. It would be a questionable business decision to rely on your revenues solely from PSA 7.5s. And lastly, I don’t see the motivation for existing TPGs to change their grading model.

While I too might be reluctant to submit a hundred-year-old ungraded card to a computerized grading system, for fear that it might not receive randomly generous treatment but instead be analyzed dispassionately (and accurately), factoring in warts and all, I’d certainly be more inclined to favor purchasing computer-graded cards over those graded by a “kid” who’s been trained in the art for a matter of weeks. 8->

Cheers,

- David.

[CardboardCollector]

Originally Posted by hilgum

Wow - so much faith in quickly trained 20-somethings, and so little faith in established technology.

The problem with computerized card scanning and grading is not a technical one—it’s a business-case one. Computer-aided visual inspection has been a part of quality control in manufacturing for decades. You’d be hard pressed to find a sophisticated manufacturing enterprise that doesn’t utilize it. And meanwhile, the folks at Google and elsewhere are developing driverless-car technology based in part, and heavily reliant on underlying edge-detection and pattern-matching algorithms. I can use Digital ICE on my flatbed scanner to remove dust and scratches detected in scanned images, and then do further image editing in Photoshop, relying on its sophisticated image-analysis tools (think Magic Wand!). My iPhone camera focuses automatically. And a lot of this is virtually instantaneous. I would argue that the only real technical problem for card grading is that you would risk ending up with a system that would find—and make grading decisions on—details that are not actually visible to the naked eye.

Look at the grading standards for the major TPGs: centering, focus, sharpness of corners, breaks in surface gloss, stains, print or refactor lines. The numerical grades simply quantify the measurements of these features. If these features can be seen—and they can!—they can be quantified and factored into an evaluation function that can be tailored to any specific subgroup of cards you want to define (machine-cut, perforated, hand-cut, … T206, ’52 Topps, ’71 Topps, etc.).

But what’s the business case for such a system tailored to sportscard grading? Like everything else, it’s expensive technology, even when a lot of the hard technological work has already been done. As Pete (ullmandds) has mentioned, most cards have already been graded. This is an interesting observation, and I think largely true (ignoring all the cards now and forever more being produced, obviously). What’s the motivation for people to submit already graded cards to a computerized TPG? Andy (bn2cardz) is also right - there isn’t any “hope” for an anomalous upgrade from a deterministic algorithmic process. Why risk obtaining a lower rating on an existing PSA- or SGC-graded card. Why spend money to have a mid-grade card boosted, when it’s generally the case that a mid-grade card is a mid-grade card for an obvious reason. It would be a questionable business decision to rely on your revenues solely from PSA 7.5s. And lastly, I don’t see the motivation for existing TPGs to change their grading model.

While I too might be reluctant to submit a hundred-year-old ungraded card to a computerized grading system, for fear that it might not receive randomly generous treatment but instead be analyzed dispassionately (and accurately), factoring in warts and all, I’d certainly be more inclined to favor purchasing computer-graded cards over those graded by a “kid” who’s been trained in the art for a matter of weeks. 8->

Cheers,

- David.

+1 Completely agree. This is not really a technology issue, it's business issue. Are you going to develop the capability and license or sell it to a TPG? What's the cost of developing the software and hardware? What's the cost of using existing technology (i.e. scanners, software, color matching, colorimeters, spectrophotometers) How many graders are employed by a TPG? How much do they make? How many cards can a grader review in a work day? What's the turnover rate of graders with a TPG? Are you going to start your own TPG?

[Eric72]

I think a machine could provide consistent grades. However, it would only be a matter of time before the scammers were able to slip fake cards past RoboGrader1000.

[parkerj33]

I would like to chime in here. I lead a software R&D team and we have some of the brightest Machine-vision software engineers around. We solve problems like these every day. We are involved with medical applications, not sportscards, but i can assure everyone here that the technology exists to identify card issue and to identify flaws (or perfections) quite easily. We do things like detect cancerous cells from normal cells via software algorithms. As many have said, the technology to detect or analyze will not be a show-stopper.

Its just a matter of effort and dedication to build the card-specific application of known algorithms. I am developing a user-friendly tool that allows non-machine-vision people to build their own algorithms out of our technology. Its possible this could be used to identify and classify cards by issue, and to also do some analysis of surface, corners, wear, scratches etc. Like someone said, this kind of stuff is done in manufacturing every day. Your iphone screen gets to you without scratches because of these types of systems in manufacturing to weed out the bad units.

I would add that detecting reprints from real could be very tough to do. the feel of card stock might be very difficult to assess. But the software could easily detect the printer "dots" if it came from a laser printer or photocopy and not a real print process.

[Bruinsfan94]

Originally Posted by parkerj33

I would like to chime in here. I lead a software R&D team and we have some of the brightest Machine-vision software engineers around. We solve problems like these every day. We are involved with medical applications, not sportscards, but i can assure everyone here that the technology exists to identify card issue and to identify flaws (or perfections) quite easily. We do things like detect cancerous cells from normal cells via software algorithms. As many have said, the technology to detect or analyze will not be a show-stopper.

Its just a matter of effort and dedication to build the card-specific application of known algorithms. I am developing a user-friendly tool that allows non-machine-vision people to build their own algorithms out of our technology. Its possible this could be used to identify and classify cards by issue, and to also do some analysis of surface, corners, wear, scratches etc. Like someone said, this kind of stuff is done in manufacturing every day. Your iphone screen gets to you without scratches because of these types of systems in manufacturing to weed out the bad units.

I would add that detecting reprints from real could be very tough to do. the feel of card stock might be very difficult to assess. But the software could easily detect the printer "dots" if it came from a laser printer or photocopy and not a real print process.

So if it could be done fast enough, the best way would be to have a live person check the authenticity, and then have the machine double check things like print lines, and grade the card? Seems like that would be great. Saying as how we are in the early days of all this technology and card grading is very small, but baseball, and some other sports are getting bigger, I bet we end up seeing something like this sooner or later. Maybe one of the companies trys to make it a high end product or something. I doubt baseball grading is a huge business but there are three companies with huge followings, so competition must be fairly fierce.

[Mdmtx]

Actually if the lowest company on the totem pole were to offer something like that it would possibly render psa to be like gai currently and devalue psa cards because people would wonder why the seller didn't reholder.

[steve_a]

It was the VCP guys that were working on this. I love the idea, especially if they did multiple sub-grades like Beckett. Been thinking about it since I saw this post. Any update Bobby?

http://www.net54baseball.com/showthr...=202183&page=4

[Peter_Spaeth]

Originally Posted by Eric72

I think a machine could provide consistent grades. However, it would only be a matter of time before the scammers were able to slip fake cards past RoboGrader1000.

Or the guy in Mexico would get his own and reprogram it.

[sportscardtheory]

I would believe that this would only be valid if done through a reputable company with their reputation on the line. Just like now, people who grade their own cards have no rep or brand name to put behind their cards, so they would be an afterthought, just like they are now. It would have to be a valid start-up or a company like PSA, SGC or BGS to get the ball rolling on this concept.

[steve B]

Interesting stuff.

I have no doubt at all that current technology could easily handle the easy things.
Size, no problem.

Corner wear also no problem. With exceptions on stuff like 88 score with three different die cuts. The first left tufts of torn cardstock on each corner, the second left those about 1/4 inch in from the corners on each edge. I'm sure that could be programmed around.

Edge wear - Tricky, but only because of the physical handling required to scan edges. Yes, even a hobby level robot could do it, but I'm not sure I'd trust a really expensive card to even a really nice commercial/industrial machine.

Identification would be possible, and probably fairly easy. The hard bit as I mentioned would be building the database. I know some of the automated systems can discern between a variety of objects on a conveyor, but as far as I know those systems are limited to picking out a known object from a group of other known objects, with maybe a few hundred different? The one I've seen only had to deal with about 10 different objects. How many cards are there? Just Topps base cards between 1970 and now is around 32000 not counting variations. And that's just base cards from one company.

Determining fakes from real based on the printing? Maybe easy, maybe not. You'd need fairly high resolution scans, I'm not sure how fast commercial scanners can do that. Certainly faster than my home scanner, but how fast? And that's where I think the data problems would get really troublesome. I would believe that adjusting for registration issues can be done simply. The same would probably be true for over/under inking, print lines smeared ink, etc. However once you're comparing dot patterns, the number of different cards goes up a lot. Some sets that had multiple print runs are different when seen at that level. Some sets have cardstock differences that are maybe challenging to tell apart from toning/staining. One of the earlier Topps Allen and Ginter sets had printed on "tobacco stains" (I learned about them when I was opening packs and eating a chocolate bar, I thought I'd ruined an entire box until I realized those stains were part of the design.

Presumably you'd need a known good example to compare the new image to.
That's a LOT of images and pretty big database. Yes, the storage isn't a problem, and a good system is fast enough that it's not really a problem. But the time to create that database and library of images with fairly high resolution seems a bit daunting to me. yeah, the junk wax era could be run through a really automated scanner, but you might have trouble getting access to the more expensive or rare cards.


Surfaces - Ok, I may have to surrender this point. I'm not positive, only because stuff like a scratch on glass usually changes the color where the scratch is, and that's easy for a machine. I'm not sure about something like fine scratches on a very glossy card, or a scratch on the back of a card where it's maybe a thousandth deep, and the same color as the surrounding cardstock, and can be similar to natural imperfections of the material.


I mentioned it poorly, others have said it better that it's more of a business case. How much do the really good systems cost for the physical machine and the development? And how fast can the cards be imaged on all 6 sides and compared? It's probably not hard to hire 3-4 people who can reasonably accurately grade a card in a few minutes for about the same as one good developer. And their hardware/technology cost is far lower, a decent magnifier is under a hundred maybe under 10 depending on how fancy you want. A top quality ruler and usable caliper under 50, and however fancy a computer you want for data entry.

Steve B

[bcornell]

[REMOVED] This forum is done.