New acsearch die match/previous sale function

I just tried out the new image search function at acsearch; it's available from their main search page. The point is to find previous auction records of your coin, and/or die matches. The results come as a list of coins, with each receiving a score out of 10, corresponding to the probability of a match. For me, the exact same coin got a score of 9, while a die match got a 4.6. I knew about the auction result already, but not the die match, which I was glad to find.

Here's the coin, a very recent auction win:



And here's the die match that acsearch found for me:

edited - copyright

Pretty cool, eh? It found the match even with the o/c reverse.

To use this new function, you just click on the image search button and upload an image, or paste in an image url. It gets processed automatically and the results are nearly instant. It's pretty slick!

Unfortunately it's not totally free, it costs 0.70 euros per successful search - if you don't get any results you pay nothing. (If you have a premium account already, you will have 20 free searches in your account.)

The only drawback I see is that if you already know of some matches - for example from the auction where you originally bought the coin! - then you'll still have to pay for the search. But that's a relatively minor drawback. Kudos to acsearch for only charging when you get results, and for keeping the price low.

Here are the usage instructions:

Since acsearch contains professional coin photographs for the most part, the best search results are obtained by using a professional coin image for your search query. Please note that:
The image should contain both sides of the coin.
The background of the image should be white (no patterns and no textures, e.g. by cropping in Photoshop, Gimp, etc.).
The image should be cropped and not contain a frame.
The shadows on the image should be as natural and unobstrusive as possible.
Snapshots from a smartphone are often too dark and blurred and, therefore, not suitable for the image search engine. In general, the better the query image, the better the results.
​

I'd be interested to see any hits that you get!

 

Log in or Sign up to hide this ad.

Nice. Coin and match. Congratulations.

 

Nice. I'm needing to do a die study on a coin I'm interested in, will have a look at this.

 

How does this stack up against ex-numis or other available tools?
I suppose there are 2 major components to its usefulness - the quality of the image matching software and the extent of the past sale database.

 

Cool!

I just signed up and found a provenance of my Alexander III Amphipolis Tetradrachm.


View attachment 970852

This exact match earned a 9.2

edited - copyright

 

Nice coin.
+ I only needed to bid $3 more to win it

 

Ooops, sorry about that @arnoldoe...

 

I thought the group would be interested in reading about my experience with the ex-numis and the acsearch image search engines.

My conclusion is that each works best for what each promotes: ex-numis for provenance, and acsearch for die links. However acsearch seems to be limited to results after 2000.

I tried ex-numis with 15 coins. I was able to exclude those results I knew. Of my 15 submissions, it found one new provenance

edited - copyright

I then tried 10 searches with acsearch using many of the same coins. Usually it picked up purchases since 2000, but nothing prior. For example, it found three listings of the above coin when it was in the market since 2000, but neither of the two times when it was auctioned prior to 2000.

I then tested how acsearch functioned finding die links with two common coins: an Athenian fifth century tetradrachm, and a first century sestertius. It returned no results for the tetradrachm and about ten probably correct die links for the sestertius.

 

Thanks @Severus Alexander for starting this thread. I used half my credits. Mostly I found matches to the auctions I bought the coins from. However...there were a couple of exceptions. I will discuss one of them here.

I did an image search for the following coin


I bought the coin from a Noble Numismatics auction in April. If you did an image search you would would find out the hammer was 105 USD plus of course the auction fees and shipping and handling.

The search I did turned up another entry. It turns out that this coin was auctioned off by H.D. Rauch in 2009. The hammer price? 333 USD plus auction fees and shipping and handling. In other words someone paid more than three times the amount for this coin ten years ago. Coin values do not always rise. I wish the photo could capture the iridescent toning...it is beautiful.

I am going to add this provenance to my notes.

Here is the Rauch image. The obverse looks the same but the reverse looks like some light cleaning was done.

 

To be fair, acsearch just responded as follows to my question about which auctions they have indexed. (Their response was received after my post.)

quote
Please note that the image search will only return results of auctions that are indexed. You can find a full list of all auctions indexed here:

https://www.acsearch.info/auctions.html

We have indexed a bit more than 100 pre-1999 catalogues already and are continuing to add more every year. You can find a complete list here:

https://www.acsearch.info/auctions.html?pre1999=2

We plan to add NFA, and about 500 other catalogues, later this summer.
unquote

 

I tried a few searches when the service was unveiled and tried it again just now using the auction house images of my Entella tetradrachm. The first hit, scoring 9.0, was the auction listing (the picture I used). Is 9.0 the highest score? I hope so since it was not just the same coin but the same image I used for the search. The second hit scored only 5.1 but it was also the same coin but from a prior auction. The exact coin made a third appearance on the list but in 8th position, scoring only 4.2. Those images were of course slightly different but at a glance it is easy to see that they are all the exact same coins. I haven't scrutinized the other 31 hits but it looks like many are obverse die matches. Those scores range from 1.1-4.5 so I'm not sure how useful this is.

While this service is nice, I'll probably just continue to search for prior sales and die matches manually. Is it worth spending money to have a computer look at the pictures for you? For instance, @Orfew could have found the prior sale of his Domitian/Minerva denarius by doing a search, although I guess with a common type of coin the large number of hits makes for a very tedious search. However, based on the plethora of Flavian rarity/variations shown by David, Andrew, and others, apparently those are frequently mis-cataloged because the cataloguer doesn't notice whatever detail makes them different. Perhaps an image-based search would help catch these mis-catalogued sales, although it is still going to require a human looking a bunch of search hits to find ones which are truly matches.

 

Once again @TIF you have hit the nail on the head. Yes, Minerva denarii for Domitian are often misattributed because of the very small differences that separate the types. (This can drive some crazy, but I quite enjoy spending a few hours tracking these down) Of course I checked Acsearch for my coin. I saw several examples. However, where the image search was well worth it was where after my search it put 2 examples one on top of the other. Sounds like a small thing but it is not. By comparing the 2 coins I was not only able to confirm that they were the same coin, but I also found out something interesting about my coin. The second image, the one from Rauch was only a 39% match. I soon found out why. It runs out that while the obverse was not touched, the reverse was cleaned slightly. This cleaning changed the look of the reverse and led to a lower percentage match. There is another reason why I like this image search. It turns out that my coin was lot 464 in the Rauch 2009 auction. Lot 465 was another example of the same coin. Lot 465 did not appear in the image search because it is from different dies.

In short, though my searches mostly turned up previous sales I already knew about, the image search led me to find out something new about my coin.

 

If I were in charge of rolling out such a service, I would never use a 100% claim just to avoid the nasty letters when there was one that was wrong. That is why some of us refuse to say a coin is real or fake based on the photos. I have a 99.9% opinion on many but have nothing to gain answering requests from people out there determined to prove the earth is flat or I am a fool. People who live in a valley have good evidence that the earth is concave.

I am not into this current fad of researching provenances to include mostly sales listings. I find it interesting that my coin was once Dattari but I could care less that it was consigned to Pegasi and CNG. IF searches were free, I would do many to learn how they work and, possibly, help the programmers improve their product. I have curiosity, for example, how it will handle things like the two photos below. The coin is Dattari/Savio and Empire Coins 1987.


Would it handle the differences in lighting and background? Would it detect that one is a cast made from the other? Could it tell the same coin before and after cleaning? Tooling? Repatination? If the sellers of this service were primarily interested in improving the product, they need to be working to improve it. Not including something like the NFA sales from the start seems questionable. Do they include that stack of poorly printed Alex Malloy catalogs I have from 30 to 60 years ago? He had many interesting coins but printed on newsprint. In those days few sales photographed every coin and lack of detail would make matches harder except for the high end coins that already are more likely to travell with papers. Can they match to Dattari Savio rubbings? To plaster casts? To published collections lost to WWII bombing. To the plates in Mouchmov showing some Reka Devnia coins (if so, they wold be stolen since they should still be in the museums). It is a great technology and will be a great resource for FurryFrogJR. I had a toy calculator once that did basic 3rd grade math. My grandson has one that preforms miracles. When image recognition is 'calculus' level, will I be around to see it?

 

Hi everyone,

thank you very much for your interest in this new tool and for sharing your experience with it (especially to @Severus Alexander, who sent me the link to this thread). I am very glad that some of you already found some provenances and die-links that they were not aware of.

Please do not pay too much attention to the score (1 to 10), which is mostly used by the system for sorting. The score does not say anything about how similar two coins (or the same coin on different images) are, but how similar the algorithm thinks the two images are. If you have two images of the same coin, for example, the score can be very low if the lighting is too different. And even if you think the picture is 100% identical, the score can be less than 10 if your query image has another format or size than the image that we've used for indexing.

This said, the same coin on different images usually has a higher score than different coins with die-links. Thus, we use it for sorting.

Regarding @dougsmit's questions:

The search can indeed handle small to medium differences in lightning and background. However, in general you will get better results if you use a white background with no texture and very natural lighting with decent shadows, as is written in the usage instructions.

It strongly depends on the quality of the cast and any possible alterations. If it's a good cast, it certainly will find it.

Again, it strongly depends on the degree of cleaning or any other alteration. I often use it before I buy a coin (especially gold) in order to see if the coin was altered (e.g. plugged holes, etc.). Here's a great example:

https://www.acsearch.info/search.html?id=5714978
https://www.acsearch.info/search.html?id=3104461

If you search for the image in the first link, you will certainly find the second entry.

 

Here is a screenshot of the example mentioned above.

edited - copyright

 

I would think it appropriate for you to add a comment to these pointing out the repair.

This reply reads as if you consider it good that the s/w matches genuine and fake. Providing a provenance path to fakes is hardly a benefit again unless the fakes are flagged in the comments. Being able to find the pre-repair coin is most certainly a benefit. I see this software as being increasing valuable as time adds refinement but now the question is whether it is more valuable in a positive sense than it is compromised by false positives/negatives. Thank you for the answers and example. I'm sure it is now a value to collectors of high end gold. I have to wonder if the buyer of that last repair still has the coin or is aware of its status.

 

I really do, indeed. Consider that most images in old auction catalogues show plaster casts and not pictures of real coins. Furthermore, the image search is simply a tool, nothing more - same as the text search. It can help you to narrow down your search, but certainly not replace a human brain and expert knowledge.

 

It seems a useful tool.

Interesting that the repaired Diocletian aureus seems not to have sold (at least the May 2016 auction listing included its previous, holey, appearance in 2014). And that the hole was back in its last appearance, when it sold for less in September 2016, than in 2014.

ATB,
Aidan.

 

I don't think we can expect acsearch to flag fakes in the comments, considering their database contains about 6 million lots. But it's good that they provide a comment box for users. Can you see my comment on the repaired coin?
https://www.acsearch.info/search.html?id=5714978

 

I can, yes.