Sunday, July 20, 2008

Microsoft Photosynth

Here is something unrelated to coins that I think is extremely cool: Microsoft Photosynth. It's not something you can buy today, but a demo from Microsoft's labs. Check out the videos!

Currently it takes Photosynth days to create a 3D model from photographs. The author's hope is to make it fast enough that a user can search for pictures of a location, feed the results into the Photosynth software, and quickly get a 3D model of the location in the photographs.

It's very impressive.

There is a paper explaining how it works here.

The first step, detailed in section 4.1 of the paper, is finding feature points in the images and matching them. The authors (Noah Snavely, Steven M. Seitz, and Richard Szeliski) use the SIFT algorithm citing D. Lowe, “Distinctive image features from scale-invariant keypoints” Int. J. of Computer Vision 60, 2, 91-110. I'd like to point out this is the same algorithm, Scale-invariant feature transform, the Image Based Recognition of Ancient Coins proposed.

That paper reported little luck identifying the kind of ancient coin in the photo, but perhaps the technique could be used to find different pictures of the same coin, in the same way the Photo Tourism paper finds different pictures of the same buildings?

If we could match a photo against a database of hundreds of thousands of coins and it was able to identify either exact matches or other coins from the same die it would be of great help to catalogers.

There are a number of interesting ways to use computer coin matching software. One way is to have a database containing scans of every coin auction catalog, 1870-1970, and match coins against it for exact matches. A success rate of even 25% would help find many coins that had lost their pre-1970 provenance. Obviously it's expensive to scan every catalog, but perhaps the ANS could persuade Google to do it for free to their library...

A second way to use computer matching is to do die-studies, finding coins in existing catalogs that share the same dies, or better yet just obverse or reverse dies.

A third way to use computer matching is to find forgeries. Coins would be checked against a database of fake coins. Getting the human out of the loop would save a lot of time for authenticators, museum professionals, and dealers. This is very important! Just recently it was discovered that a metal detectorist at a competition in England 'found' a modern Bulgarian aureus forgery (here and here).

No comments: