In the context of the article “Low-complexity arrays of patch signatures for efficient ancient coin retrieval”, we created two data sets of ancient coins which are described below.
The process of creating artificial coins tries to emulate the one used for ancient coins: a die, carved by an artist, is struck against a metal flan on which it marks its relief. A single die is used a certain amount of times, wears out and is replaced by a new one. This implementation yields several types of coins, for each of which several dies have been used. Using the protocol described in the article, a series of 16 types of coins were generated. Each type is different and is represented essentially with a large symbol at the center with or without a ring of small symbols. There is no distinction between obverse and reverse. For each type, the process designs 5 dies and 5 coins are made out of each die. The resulting coins are constructed as height maps. The Python source code to generate the height maps can be downloaded below.
We present a data set of 23 coins. This data set contains a variety of coins which can be clustered into classes and which contains coins struck with the same die. One die is used for one side and both sides of a coin may not have been struck from the same pair of dies. We can therefore consider that the data set has 46 coins, composed of some obverses and some reverses. Eight acquisitions are done on each coin (for the obverse and reverse respectively) using eight light source positions around the object.
Florian Lardeux, Petra Gomez-Krämer, Sylvain Marchand. Low-complexity arrays of patch signature for efficient ancient coin retrieval. Pattern Analysis and Applications, 27(3), 2024.
If you use one of these data sets, please cite this reference.
This work was financed by the Region of Nouvelle-Aquitaine, France.