Coin-Recognition Technology?

Discussion in 'Coin Chat' started by JCro57, Feb 25, 2018.

  1. juris klavins

    juris klavins Well-Known Member

    Neal likes this.
  2. Avatar

    Guest User Guest



    to hide this ad.
  3. Neal

    Neal Active Member

    Thanks! I had not found these sites before!
     
  4. physics-fan3.14

    physics-fan3.14 You got any more of them.... prooflikes?

    Same way they do diamonds. Every diamond certified has a tiny number engraved on the girdle which corresponds to the certificate. You can't see it unless you're looking for it, and its completely hidden. You could very easily do the same thing on the edge, especially hiding it in the reeding for applicable coins. Now, I don't recommend this of course, but it answers your question.

    As for the computer imaging - it would do well for recognizing coins which had been submitted, and it would help with identifying counterfeits quickly against known examples.

    However, the next progression is "Well, why can't we just have the computer grade the coin?" @cpm9ball has already mentioned it in this thread, although everyone passed by his comment. There has also been a TPG that attempted to use this method (albeit 20 years ago, before technology was advanced as it is).

    The problem is, a computer can never evaluate the eye appeal, or the attractiveness of the luster. A computer can count the number of marks, a computer can use a meter to gauge the quality of luster, a computer can evaluate the strike as compared to a perfect strike (it can even account for die marriage differences.) A computer can get every single attribution correct every time, knowing every die crack and star position and whatnot.

    What a computer can never do, however, is synergize all this into a grade as we understand it today. Remember, we market grade - we grade on a curve, based on eye appeal. There is a subjective quality to grading that a computer will never understand - it's like art appreciation. You can program a computer to analyze brush strokes and color, but it will never understand why the Mona Lisa is so amazing, or understand why Van Gogh moves people to tears. It will never understand why a magnificently toned Battle Creek is worth more, and is thus graded higher, than an average toned Morgan.

    That is why there is a certain amount of science to grading, but there is also a significant amount of art to it. And that is why a computer will never be able to accurately grade coins according to our current standards and scheme.
     
    kaparthy, ldhair, wxcoin and 2 others like this.
  5. JCro57

    JCro57 Making Errors Great Again

    Excellent on all points. Nicely presented
     
  6. Neal

    Neal Active Member

    I like your reply, but I'm not sure I'd say "never." If the computer guru's are right, in the future not only may AI be able to judge eye appeal, beauty, and art (it is already composing music in the style of any classical composer you choose), but computers may even be collecting coins themselves as relics of our civilization! (I hope not.)
     
  7. baseball21

    baseball21 Well-Known Member

    Skynet, and yes eventually we will be at that point.
     
  8. JCro57

    JCro57 Making Errors Great Again

    It might be able to judge the programmer's idea of beauty, but not for other people's ideal of what it is. For example, some people hate toning, where a programmer may love It.
     
  9. TheFinn

    TheFinn Well-Known Member

    TPG Pop Reports, like Mintage Reports, need to be taken with a grain of salt. When the U.S. Mint melted all of those bags of brand new silver dollars before WWI, did they adjust the mintage reports? When all of the recalled gold came in to be melted into bars to be put in Fort Knox, did someone write down how many 1927-D St Gaudens $20 went in each furnace charge? When private refineries were melting silver U.S. coins, did they send a note to the U.S. Treasury department to let them know how many 1944-D Walking Liberty Halves, and how many 1945-S Mercury Dimes they melted? Or how many 1964-D Peace Dollars were actually in the 316,076 silver dollars melted?
     
    wxcoin likes this.
  10. Neal

    Neal Active Member

    That is also a problem with TPG graders. It comes down to both a science (adherence to written standards) and art (just how much does the nick in that spot hurt eye appeal?). As for the programmer, with true AI the program "learns" from input data, "experience" if you will, and how it makes its judgments is often a mystery to the programmer. By inputting a large number of coins with values derived from professional graders, the program would learn, with varying degrees of success depending on how good the program was and how much data was entered, to judge grades that reflected the views of the original graders rather than the programmer.
     
  11. cpm9ball

    cpm9ball Cannot Re-Member

    I agree with everything you say, Jason, but when I first brought this subject up in a discussion on the NGC forums about 12-13 years ago, I suggested that the technical aspects you cited could be performed by the computer while the subjective aspects could be performed by humans much like the finalizers, today, being used to double-check the work of the graders.

    Chris
     
  12. baseball21

    baseball21 Well-Known Member

    There’s no point in the computer grading it at all then since you’re still using humans. Plus you would very likely be using inferior graders as the best ones aren’t going to stick around after being told that they can’t grade well enough and computers can grade better than them and be fine with only grading part of the coin.
     
  13. Jack D. Young

    Jack D. Young Well-Known Member

    PCGS's Gold Shield service is interesting in this discussion: https://www.pcgs.com/goldshield.

    I recently resubmitted an 1803 large cent on a "reconsideration" after it had originally been returned body bagged as "Authenticity Unverifiable". I had a second opinion documented on its authenticity and submitted the documentation including my own in-hand review notes with the coin. It took several weeks to get everything pulled together and PCGS sent me several notes through the process and agreed to slab it as genuine the second time, surprising me with the gold shield service and images at no cost to me- surly this took more time than they normally have to spend on a submitted coin like this.

    PCGS_Label.jpg s-260-ds.jpg
     
  14. cpm9ball

    cpm9ball Cannot Re-Member

    The point is that the computer could perform the same tasks with greater speed and accuracy without any variations. The same principles were used when the automotive industry converted to a computerized assembly line.

    I'm not suggesting that humans couldn't perform the tasks, but at least the computers wouldn't have to take coffee breaks or receive paid vacations and sick leave.

    Chris
     
  15. physics-fan3.14

    physics-fan3.14 You got any more of them.... prooflikes?

    But if you can hire a guy with some experience and knowledge that can already do it in 20 seconds with a very high degree of accuracy, do you really need to pay $5 million for some new super-wham-a-dyne system? You aren't saving anything on the grading cost, and you aren't saving time - the graders aren't the hold up in the 6 week grading process. It's the whole system and paperwork that's the hold-up.
     
  16. messydesk

    messydesk Well-Known Member

    Busy thread, and I'm going to make it busier. A couple things I want to touch on here:

    Yes and no. As stated earlier, PCGS has this with their Secure Plus/Gold Shield service. They can easily recognize if a coin has already been through for grading if they choose to scan it when resubmitted, and the grading history becomes an influencing factor in the grading process. It's not a uniform technology or database, since each TPG wants to maintain a competitive edge over the others.

    Yes, but this number would probably be easily removed from a coin without affecting value. Not so for a diamond.

    This is correct. The reason that a computer or AI can't evaluate eye appeal is that we can't arrive at a ground truth evaluation of it. Ask 9 different experts about the eye appeal of a coin and you'll get 10 different answers. A computer will do no better than that, and that's if the training and test data (lots of it) are all good.

    One interesting things about AI is that if you're able to think about how you'd develop the skill and execute the task, the AI version is probably similar. So for assessing eye appeal, I look at a coin under a consistent light, both sides, and make a judgment based on my internal "eye appeal rules," which are guided by my tastes in coins. I could look at 10000 coins and judge their eye appeal, and perhaps do so consistently (I have to if it is to be useful training data). Now think of what has to be fed to a computer to train it. We need dynamic (not static) pictures of the coins appearance to a pair of eyes (not a single eye) under consistent light that gives the same views I am privy to when I'm assessing eye appeal. Fortunately, for motion and eye appeal, I don't need huge images. 1K x 1K will do, but I need color, so 3 MB per frame. The dynamic requirement now means that for each coin I need to have many frames. If I want to be consistent from coin to coin, then I need the same views, which I don't really know in advance of seeing the coin, so figure 100 views of the coin to simulate moving the coin around. Double that, since I need both sides, and double it again, since I need to use two eyes at once to view the coin. For those keeping score, that's 1.2 GB of data for each coin. While compressed storage will shrink that, the amount of data that we're training with is still 1.2 GB once it's read from storage. If I viewed 10000 coins, we want our AI system to have that many for a combination of training, testing, and validation. 12 Terabytes, and that's just to get my opinion automated, assuming my tastes don't change, since if they do, we have to start the training step over again. And if my tastes aren't the same as someone else's the results will be worthless. It should be apparent by now that attempting to assess eye appeal, or other predominantly subjective tasks for which we can't determine the ground truth, with AI is not the hill to die on.

    One thing I really think AI could do well is detect acceptable toning. I'm avoiding the word "artificial" because I don't want to revisit the debate of exactly what artificial means. PCGS already has a ton of images of acceptable and questionable coins, and they usually take these pictures so as to show a complete picture of the toning, suppressing the luster. Assuming it isn't necessary to reassess some of them for correctness, an AI system could be trained to tell the difference between the two classifications, then classify images of test coins acquired in the same manner (suppressed luster, complete toning).
     
    ldhair likes this.
  17. wxcoin

    wxcoin Getting no respect for 63 years

    In a perfect world one would expect the software to recognize just about every possible die state. The three cent nickel type which I've taken an interest in contains many clashed die examples in many different clashed states. One would expect the software not to lower a grade because of them. A positive would be for software to uncover new varieties since an algorithm could be run through a particular coin series and date, looking for common die markers. The possibilities could be endless. But then again, that would take some of the fun out of doing the work oneself.
     
  18. -jeffB

    -jeffB Greshams LEO Supporter

    That used to be a lot of space. :)
     
  19. -jeffB

    -jeffB Greshams LEO Supporter

    Based on what we're seeing from AI practice (not "research"), I'm not convinced of that.

    Sure, any AI system will produce judgements that a human would disagree with -- but the same is true for any human expert in this realm. Eye appeal is subjective, and opinions will differ.

    I'm pretty confident that a well-trained "A-eye-appeal" system could produce results that a panel of human experts agree with at least as reliably as any one human expert could. I'll go beyond that ans say that the AI could do better than any human expert. But there's going to be a LOT of human resistance to any such effort, and a lot of people will point to any AI judgement they disagree with as a sign that the system is a failure.
     
  20. messydesk

    messydesk Well-Known Member

    While it's no longer a lot of space in terms of what a data center can hold, it is still a lot of data that needs to be collected.
     
  21. Hommer

    Hommer Curator of Semi Precious Coinage

    I too have thought of this a great deal through the years. An AI system would work and just as a weather computer can predict with a certain amount of accuracy, the system could predict eye appeal. This accuracy was/is learned by repetition. My first thoughts of this came along about the time as Coin Star machines and I wondered if they could be programmed to detect valuable coins.
     
Draft saved Draft deleted

Share This Page