Monday, July 17, 2017

Database of Digits Extracted from (modern) Coins

Xingyu Pan and imageLaure Tougne have published a paper describing their database of digits extracted from scans of modern coins.

The database of 3000+ images itself cannot be directly downloaded. They explain here that you may request it from them.

The paper is good! The databases discussed are

  • 606 digit images manually cropped from high-resolution photographs of well-conserved coins
  • 1,200 digit images automatically extracted from a subset of a PCGS database containing coins in different states of preservation
  • 1,200 digit images manually extracted from the same coin photographs as above

Although the new dataset only contains digits from modern coins the authors discuss the literature of recognition of ancient coin lettering. There are as yet no public databases of images of the digit or letters extracted from images of ancient coins.

Saturday, July 08, 2017

Antiquities Memorandum of Understanding with Libya

Please read Cultural Property Observer and comment on the Libyan MOU. If the MOU is completed and includes coins it will become difficult for Americans to buy ancient coins from Kyrene in European auctions. It will also be difficult to import Ottoman coins from the territory of modern Libya.

Do it this weekend or it will be too late.

Thursday, July 06, 2017

Recognizing Roman coins with neural networks

Jongpil Kim of Rutgers university reports a neural network capable of correctly recognizing the emperor 73% of the time from coin images. The reverse type is recognized 67% of the time.

The report is available on SEQAM and the paper is available by clicking PDF on this Cornell library site.

Wednesday, April 26, 2017

Tuesday, March 07, 2017

Video on creating 'coins' with CNC machine

I stumbled across this 9 minute video which shows the process of creating a medallion with realistic bust using a CNC machine.

The medallion produced looks quite good! It looks better than most of the 'coins' featured in do-it-yourself coin making videos. Yet most of the steps made me cringe because I was certain he was doing it the wrong way.

Once he has a 3D model on the computer, he feeds a block of metal to a CNC machine rated only for wood. I was sure something would go wrong but he got it perfect.

To produce the round shape he just takes his metal block and starts hitting it with a belt sander! I was certain he was doomed but it looked very nice when complete.

To give the appearance of patina he uses spray paint. I was amazed that this worked at all! He got a nice effect in the end.

The final medallion suffered a bit of a 'staircase effect' on the face due to the limited depth control. I would have liked to see if he could have removed that with a bit of polishing. Even with the depth artifacts he managed to produce a pretty nice medallion using techniques I would not have expected to be effective.

Saturday, February 25, 2017

Print on demand catalog of the coins of Troas

I recently purchased a Print on Demand copy of Warwick Wroths 1894 catalog of the British Museum collection Catalogue of the Greek Coins of Troas, Aeolis, and Lesbos.

Google scanned three copies of this book: from the University of Michigan, from the New York Public Library, and from Harvard

The Harvard copy doesn’t seem to show up in an ordinary Google Book search. I found it using this link:

Public domain books that have been scanned by Google can be printed on demand. To order online, just bring up the book in Google Books, click “Get this book in print” and “On Demand Books” and several vendors appear. I wanted to see an Espresso Book Machine in action. There is one at the McNally Jackson bookstore in Manhattan. So I went down there, armed with the title.

The machines aren’t self serve. I gave the operator my title, and he brought up the University of Michigan Copy. I had hoped to see it printed but the machine apparently takes a while to download the book. It downloaded overnight and the operator printed it the next morning. I was surprised to find I had received Harvard’s copy.

The problem with getting a different copy than expected is that Google’s 2008-9 analysis did a poor job of distinguishing photo plates from line art. For numismatic books this is a problem. If the Google book preview shows a coin “posterized” the print on demand copies print it as a black blob. Luckily Harvard's copy is better than the U of M's.

Here are three copies of plate 8. On Harvard and the NYPL’s copy, coins #1, #2, #18, #19 and the reverse of #22 are shown greyscale. On the U of M’s copy, none of the coins are shown greyscale. They all appear as “posterized” four color art.

The “posterized” coins appear as black blobs when the Book Machine prints them. I am not certain this is the fault of the book machine. It is unfortunate that there is no way to supply feedback to Google books that a page or part of a page is greyscale artwork.

There is also no way to specify that the printed size of the page is meaningful. This is not a huge problem for the BMC volumes, which originally appeared the similar in size to what the Espresso Book Machine prints.

I paid $17 plus tax for this volume. For comparison, the hardcover decent plate Italian edition by Forni from the 1960s can be found used for $50, with other copies selling for more. The original version can sometimes be found at auction for $150+.

Many years ago I wanted to get folks with original BMC volumes to scan the plates. I had hoped to assemble a plates-only PDF. I managed to scan a few volumes myself but was not able to get the needed volumes and the project fell through. Perhaps with the British Museum putting their collection online the need for this has passed?