Stamp Community Family of Web Sites
Thousands of stamps, consistently graded, competitively priced and hundreds of in-depth blog posts to read
Stamp Community Forum
 
Username:
Password:
Save Password
Forgot your Password?

Welcome Guest! Need help? Got a question? Inherit some stamps?
Our stamp forum is completely free! Register Now!

Problem With Expertizing; #28B Example

Previous Page | Next Page    
 
To participate in the forum you must log in or register.
Author Previous Topic Topic Next Topic
Page: of 6
Pillar Of The Community
Learn More...
4046 Posts
Posted 01/21/2020   09:41 am  Show Profile Bookmark this reply Add rogdcam to your friends list  Get a Link to this Reply
It is coming Rev. The "don't get your hopes up" refrain has been continually repeated throughout modern history when it comes to technology and then the reality exceeds the "hopes". I see it happening sooner rather than later. You make a submission. There are no employees other than a support staff or two. The equipment handles the stamp. It analyses the patient within minutes, spits out a cert, enters the information into a database, processes payment and prepares it for shipment back to the submitter. You can have an electronic cert or a paper cert with all of the data acquired in the analysis embedded. You can also request a credit card type cert. Within the cert are contained the magnified views acquired of particular relevant areas, front and back images, ink analysis data, centering data, gum analysis and so forth. Instead of a months long process it is done within days.



Send note to Staff  Go to Top of Page
Valued Member
United States
331 Posts
Posted 01/21/2020   10:09 am  Show Profile Bookmark this reply Add m and m to your friends list  Get a Link to this Reply
Don I think that you should approach the various certifying bodies and the APS with your ideas. Sending a copy of these discussions along would not hurt. The larger the groundswell the better the chance for change.
As to the comment by rogdcam re taking the expert out of the equation I hope that day never comes about. We do not need a Robbie Hal or even a Watson telling us what we have. Time and the environment have changed a lot of things, and will continue to do so despite our best efforts. No one alive today knows what the original formulas shades gums or quality controls were in place when the items were created. Even lighting has a role.
That said its still a hobby and one should indulge in it as deeply as one wants and enjoy what ever pleasures it returns.
Send note to Staff  Go to Top of Page
Pillar Of The Community
Learn More...
4046 Posts
Posted 01/21/2020   10:23 am  Show Profile Bookmark this reply Add rogdcam to your friends list  Get a Link to this Reply
.....little did the Forum know that rogdcam was a robot....

My wife is always telling me that my intelligence is artificial.
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
1342 Posts
Posted 01/21/2020   10:25 am  Show Profile Bookmark this reply Add cjpalermo1964 to your friends list  Get a Link to this Reply
rogdcam is correct that machine generated certificates are possible, with the main challenge being creating a training data set that is large enough to generate output with a confidence level that is high enough. I have spent most of the past two years writing patents for various inventions derived from artificial intelligence and machine learning. While there are periodic improvements in the underlying statistical math for prediction, most of it is well settled; the key challenge is getting the data.

A benefit of machine-generated certificates would be, if the system is programmed as I would prefer, always outputting the confidence level of the machine classifier, because this is a value that the algorithms inherently and necessarily generate. It would be materially different to have a certificate that says "REPERFORATED AT RIGHT; CONFIDENCE LEVEL = 78" rather than just "REPERFORATED AT RIGHT" as we have now. For the latter, one can always ask, who sez, and how sure were they?
Send note to Staff  Go to Top of Page
Valued Member
United States
31 Posts
Posted 01/21/2020   10:40 am  Show Profile Bookmark this reply Add widglo46 to your friends list  Get a Link to this Reply
I think Rodgcam is right about the possibilities of AI taking over stamp identification. The only thing holding it back will be the expense of implementing it. The technology is already available today. Computers can read chest X-rays more reliably than radiologists. Another example - utilizing neural networks, they can determine the sex of an individual from their retinal scan with 99% accuracy. What is mind-boggling for me is the fact that neither retinal specialists nor computer scientists can explain how they do it. It is like trying to understand how the brain works.

I also agree with 51studebaker - "... when an organization addresses their mistakes and makes improvements they are being managed correctly and increases the value of their products and services." . I am basically an unknown in this hobby, and my opinions don't carry the weight of the "Pillars". Perhaps the PSE will get the message if more of you approach them.
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
1307 Posts
Posted 01/21/2020   11:14 am  Show Profile Check cfrphoto's eBay Listings Bookmark this reply Add cfrphoto to your friends list  Get a Link to this Reply
It would be a stretch to assume that any of the major expertizing committees has a certificate database. While each committee may have imported a list of certificates with some descriptive attributes into a database table, it is doubtful that much was done other than to add indexes on columns like certificate number, certificate date, country or catalog number. Ability to query and access data also depends on software than can query and present results to users. The Philatelic Foundation was able to make use of web and database software like Siegel Auctions and the United States Philatelic Classics Society. The American Philatelic Society front end provides access to the list of certificates and images originally developed to internally track certificates. At the time, no thought was given to capturing expertizer comments from the paper certificates. Absent an on-line expertizing form to capture expertizer comments, the best that could be managed would be to capture readable images of every paper certificate form. Because of the cost, this data is likely to remain moldering in metal file cabinets. PSE certificates can be searched by certificate number. Is PSAG certificate information available on the Internet? What about society or foreign certificaets?

Automating some parts of the certification process may be possible, but not enough examples of the rarest stamps exist to establish rules. Instead, tracking every known example like in Siegel population reports would be helpful. Rules of thumb used by expertizers would also need capturing and validation. For example, Design dimensions of 1875 reissues differ slightly from the regular issue 1869 US Pictorial issue. The description of size differences in the Philatelic Foundation Opinions article is not consistent. Resolving this type of issue could involve building an intelligence database capable of capturing conflicting data, opinions about the data and adding links between related data fragments with each section of the database managed by a subject matter expert to determine which version of the data is preferred or most probable. While public domain software exists, the resources required may not be available. This is the same issue faced by the editors of the Scott Catalog. Without a database that can be used to produce reports (catalogs), they still stuck in the linotype hot lead era, unable to make changes and stay consistent without extraordinary effort.
Send note to Staff  Go to Top of Page
Valued Member
United States
216 Posts
Posted 01/21/2020   11:49 am  Show Profile Bookmark this reply Add Bud to your friends list  Get a Link to this Reply
Even without full automation or new databases, it ought to be practical for expertising services to implement cjpalermo's excellent suggestion to include a confidence level assessment in a cert. Without this kind of measurement, in my opinion most certs aren't worth the paper they're printed on.
Send note to Staff  Go to Top of Page
Moderator
Learn More...
7727 Posts
Posted 01/21/2020   12:14 pm  Show Profile Check 51studebaker's eBay Listings Bookmark this reply Add 51studebaker to your friends list  Get a Link to this Reply
Algorithms for determining the centering on stamps already exists and is not rocket science, it is easy to quantify. Image recognition also already exists as well as lookups for stamp matching to existing databases.

But there are still a number of challenging aspects which would need significant amounts of work before being ready for primetime use in a certification environment. Cancels offer both problems and solutions; they can make recognition difficult by obliviating key parts of the stamp and but they can also be helpful with dates and locations. Software to deal with these things would be complex and a significant body of work.

And stamp colors would also be problematic until non-destructive molecular/atomic level analysis or the paper and ink is widely available.

Stamp condition and altered stamps is another 'mixed bag'; some things would be easy to code (pulled perfs, visible ceases, tears, etc.) but other would be difficult (rebacked stamps, expertly filled thinks). Reperfs and all perforations calculations are straight forward and not difficult to code. There are also several subjective attributes which present problems. 'Freshness' of a stamp, color alignment in multiple pass printings come to mind. Watermarks are also challenging although some of this may be gleaned from other 3rd party technologies software (i.e VSC8000).

Adding bar codes is a no-brainer, anytime anything paper is printed there should be bar code(s) on it. It is free and can be as simple as just containing a link back to the certification organization website. But obviously it also lends itself to containing all the data on the front of the cert so that in the future it is possible to scan the document and have digitized data. In this day and age when I see a philatelic document or publication without a bar code I think am dumbfounded. Business cards, flyers, handouts, marketing cut sheets, books, magazines, etc should all have bard codes. The cost of printing a bar code virtually zero.
Don


Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
1342 Posts
Posted 01/21/2020   12:26 pm  Show Profile Bookmark this reply Add cjpalermo1964 to your friends list  Get a Link to this Reply
The bottom line is that the training data would be a small dataset and every record would require manual labeling for ID purposes by someone like Clark, so while the solution is technically feasible, it is probably economically infeasible, even with free crowd-sourcing. There are too many stamps, too many variants and not enough expert hours.
Send note to Staff  Go to Top of Page
Moderator
Learn More...
7727 Posts
Posted 01/21/2020   12:36 pm  Show Profile Check 51studebaker's eBay Listings Bookmark this reply Add 51studebaker to your friends list  Get a Link to this Reply
Agreed, but implementing a certification SQL database is not nearly as challenging as automating a stamp certification process. Manual data entry is no more time consuming than filling out a form by hand or using a typewriter/computer to generate a cert. Maintaining relationships between certs in a SQL DB is also straight forward and could be done by virtually anyone with minimum SQL experience.
Don
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
508 Posts
Posted 01/21/2020   12:51 pm  Show Profile Bookmark this reply Add funcitypapa to your friends list  Get a Link to this Reply
More recent PSE certificates are on line but the earlier ones in the format listing the expertizers as exemplified in the 28b cert from 1997 are not—and I am not quite sure why. Apex, the oldest US expertizing committee only shows certs going back a couple of years
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
6326 Posts
Posted 01/21/2020   8:47 pm  Show Profile Bookmark this reply Add revcollector to your friends list  Get a Link to this Reply
I can't wait to see all the data on reperfing. "REPERFORATED AT RIGHT; CONFIDENCE LEVEL = 78" sure makes me want to see it. NOT.
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
2116 Posts
Posted 01/21/2020   10:35 pm  Show Profile Check eyeonwall's eBay Listings Bookmark this reply Add eyeonwall to your friends list  Get a Link to this Reply
I agree with cjp that AI for expertzing might be possible but probably not economically practical at this time.
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
2191 Posts
Posted 01/22/2020   08:32 am  Show Profile Check sinclair2010's eBay Listings Bookmark this reply Add sinclair2010 to your friends list  Get a Link to this Reply
If we think we have problems now with expertization, wait for computers to do it.

While I agree that the OP's stamp has been reperforated at right, it is also at least partly reperforated at top. I can say that with a very high level of confidence, like 99.9 percent...
Send note to Staff  Go to Top of Page
Pillar Of The Community
United States
6326 Posts
Posted 01/22/2020   12:46 pm  Show Profile Bookmark this reply Add revcollector to your friends list  Get a Link to this Reply
"Reperfs and all perforations calculations are straight forward and not difficult to code".

I suggest that perhaps you have not spent quite enough time CLOSELY examining perforations, both genuine and fake. The size and shape variations on genuine perfs are infinite. The easy reperfs are not the problem now, and will not be then. But the really difficult ones will also have a very wide range, and cannot come down to simple mathematics.
Send note to Staff  Go to Top of Page
Page: of 6 Previous Topic Topic Next Topic  
Previous Page | Next Page
 
To participate in the forum you must log in or register.


Go to Top of Page
Disclaimer: While a tremendous amount of effort goes into ensuring the accuracy of the information contained in this site, Stamp Community assumes no liability for errors. Copyright 2005 - 2020 Stamp Community Family - All rights reserved worldwide. Use of any images or content on this website without prior written permission of Stamp Community or the original lender is strictly prohibited.
Privacy Policy / Terms of Use    Advertise Here
Stamp Community Forum © 2007 - 2020 Stamp Community Forums
It took 0.39 seconds to lick this stamp. Powered By: Snitz Forums 2000 Version 3.4.05