Options
The PCGS Expert
foodude
Posts: 3,591 ✭✭✭
The following is a reply I wrote for another thread titled "Totally Impractical Idea For PCGS". I am repeating it here since others may not have stumbled upon it under the other thread, and I thought people that are newer to the hobby may not be aware of it and find the story interesting.
PCGS had a system they dub "The PCGS Expert" (a machine for grading coins) that could "fingerprint” if you will a coin by digitally capturing various aspects of a coin. The Expert is said to have been programmed with human defined methods and simulates the human visual system.
"The PCGS Expert" (photos attached) was developed in the late 80's, and announced along with a description and photos of the system in June, 1990. The "Expert", according to a June 5, 1990 PCGS ad in NN, consisted of a Macintosh IIX, using a 50 megahertz Motorola 68030 with a floating point co-processor, high speed cache, eight megabyte memory, erasable optical laser compact disk, video capture hardware, high precision robotics, and an attached special purpose processor.
The “Expert was designed to grade Morgan dollars, with the thought that in the future it could programmed for other series such as Walkers, and $20 St. Gs. The plan was for the Expert to determine a grade, then a human(s) would verify the grade. They tested it with more than 10,000 coins.
Here is how the process worked for grading a Morgan dollar. In step 1, multiple images of the coin under various lighting conditions were captured in digital form using a high resolution camera. In step 2, all or various portions of the captured image were enhanced to bring out important features of the coin. In step 3, the key regions of the coin were examined in great detail to identify, classify, measure, and score all flaws. In step 4, secondary regions of the coin were examined to identify flaws that existed in busy background regions such as the hair, letters, and rim. These flaws were then classified, measured, and scored. In step 5, a light flow and reflectance analysis was used to precisely measure the mirror as well as the inherent luster of the coin. In step 6, the key areas of the coin were examined to measure the strength of the strike including the hair, particularly the hair above the obverse ear. In step 7, dozens of aspects of the coin were examined to define measures such as satin, smoothness of light flow, “flash”, color, and toning to establish the mood and eye appeal of the coin. In step 8, thousands of parameters were generated from the various analyses and these were then synthesized into key components of the coin including obverse and reverse marks, strike,. Luster, eye appeal, mirror, toning, key metric areas, and exceptional toning. Finally, in step 9, the results of the synthesis were then combined using a large set of “expert rules” to establish the final grade for the coin. The final grade of the coin was then verified by a human(s).
I believe the Expert was subsequently shelved. I don’t recall why, but I probably have the answer somewhere in my collection of notes.
Here’s some additional info: On February 22, 1991 PCGS entitled “Automated coin Grading System” which subsequently led to two US patents- US Pat. Nos. 5,220,614 and 5,224,176. The patents are assigned to PCGS.
Here are the websites for copies of the patents, if you really want technical details. I have not reviewed them as yet, but I suspect they either further describe the "Expert" system or a variation thereof.
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1='professional+coin+grading+service'.ASNM.&OS=AN/"professional+coin+grading+service"&RS=AN/"professional+coin+grading+service"
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=2&f=G&l=50&co1=AND&d=ptxt&s1='professional+coin+grading+service'.ASNM.&OS=AN/"professional+coin+grading+service"&RS=AN/"professional+coin+grading+service"
Can someone post these links properly. I can’t figure out how to do it without typing the links in. Isn’t there a way to copy the link into the correct place?- If you know PM me.
PCGS had a system they dub "The PCGS Expert" (a machine for grading coins) that could "fingerprint” if you will a coin by digitally capturing various aspects of a coin. The Expert is said to have been programmed with human defined methods and simulates the human visual system.
"The PCGS Expert" (photos attached) was developed in the late 80's, and announced along with a description and photos of the system in June, 1990. The "Expert", according to a June 5, 1990 PCGS ad in NN, consisted of a Macintosh IIX, using a 50 megahertz Motorola 68030 with a floating point co-processor, high speed cache, eight megabyte memory, erasable optical laser compact disk, video capture hardware, high precision robotics, and an attached special purpose processor.
The “Expert was designed to grade Morgan dollars, with the thought that in the future it could programmed for other series such as Walkers, and $20 St. Gs. The plan was for the Expert to determine a grade, then a human(s) would verify the grade. They tested it with more than 10,000 coins.
Here is how the process worked for grading a Morgan dollar. In step 1, multiple images of the coin under various lighting conditions were captured in digital form using a high resolution camera. In step 2, all or various portions of the captured image were enhanced to bring out important features of the coin. In step 3, the key regions of the coin were examined in great detail to identify, classify, measure, and score all flaws. In step 4, secondary regions of the coin were examined to identify flaws that existed in busy background regions such as the hair, letters, and rim. These flaws were then classified, measured, and scored. In step 5, a light flow and reflectance analysis was used to precisely measure the mirror as well as the inherent luster of the coin. In step 6, the key areas of the coin were examined to measure the strength of the strike including the hair, particularly the hair above the obverse ear. In step 7, dozens of aspects of the coin were examined to define measures such as satin, smoothness of light flow, “flash”, color, and toning to establish the mood and eye appeal of the coin. In step 8, thousands of parameters were generated from the various analyses and these were then synthesized into key components of the coin including obverse and reverse marks, strike,. Luster, eye appeal, mirror, toning, key metric areas, and exceptional toning. Finally, in step 9, the results of the synthesis were then combined using a large set of “expert rules” to establish the final grade for the coin. The final grade of the coin was then verified by a human(s).
I believe the Expert was subsequently shelved. I don’t recall why, but I probably have the answer somewhere in my collection of notes.
Here’s some additional info: On February 22, 1991 PCGS entitled “Automated coin Grading System” which subsequently led to two US patents- US Pat. Nos. 5,220,614 and 5,224,176. The patents are assigned to PCGS.
Here are the websites for copies of the patents, if you really want technical details. I have not reviewed them as yet, but I suspect they either further describe the "Expert" system or a variation thereof.
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1='professional+coin+grading+service'.ASNM.&OS=AN/"professional+coin+grading+service"&RS=AN/"professional+coin+grading+service"
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=2&f=G&l=50&co1=AND&d=ptxt&s1='professional+coin+grading+service'.ASNM.&OS=AN/"professional+coin+grading+service"&RS=AN/"professional+coin+grading+service"
Can someone post these links properly. I can’t figure out how to do it without typing the links in. Isn’t there a way to copy the link into the correct place?- If you know PM me.
Greg Allen Coins, LLC Show Schedule: https://forums.collectors.com/discussion/573044/our-show-schedule-updated-10-2-16 Authorized dealer for NGC, PCGS, CAC, and QA. Member of PNG, RTT (Founding Platinum Member), FUN, MSNS, and NCBA (formerly ICTA); Life Member of ANA and CSNS. NCBA Board member. "GA3" on CCE.
0
Comments
<< <i>I believe the Expert was subsequently shelved. I don’t recall why, but I probably have the answer somewhere in my collection of notes. >>
My guess it was because it took more than the alotted six seconds per coin.
Very cool info, BTW!
Russ, NCNE
not be cost-effective.
The human eye and high magnification is the best way so far.
Sure the computer could read a coin but it still would take someone to interpret the data and check for
some calibrated accuracy indicators.
I think it would work great for SQs since the toning issue wouldnt be a factor.
It could spit out more than 100 a minute if a computer did it.
Russ, NCNE
When they were using the "Expert", a few dealers who were paying attention, quickly figured out that the way the machine was programmed, it graded the better date Dollars a bit differently than the human graders did. (as in looser) For whatever reason, it was easy to get lackluster, but mark-free P and O mint coins graded 64 and 65. Also, it was easy to get REALLY blasty better date S mints with bad marks in the fields, particularly on the back to grade quite high. Other things that the "Expert" really liked were coins that had a "washed out" look from being dipped too many times and it loved CC mints because of the luster. The "Expert" really disliked bagmarks on the face, but didn't seem to mind weakly struck coins. As for being checked by a human, I would have to guess that after looking at enough coins, the person checking the machines work just started to believe that the machine was right and let the coins through.
All in all, it was a pretty good idea, but as they say in the programming game, garbage in, garbage out. In this case, it is my belief that the machine saw all of the dates more or less as equals, which certaintly isn't the case when humans grade them. Perhaps with a larger sampling, say a million coins, this would work. It's kind of like Deep Blue. Before IBM spent millions on programming, a human could play better chess than a computer. Perhaps with a few million in programming, a computer could grade coins better also.
I'm willing to bet that David probably won't want to find out.......