O.S.C.A.R.

The O.S.C.A.R. detection system ("Optical System for Coin Analysis and Recognition") was specially developed by the Fraunhofer IFF in cooperation with the LDA in 2017 during the first digitization project Digital Heritage.

Flat objects such as coins, medals or medallions made of gold, silver or bronze with a diameter of 5 to a maximum of 75mm can be used with O.S.C.A.R. and recorded digitally. At first it doesn't matter whether they are fragments, halflings or bent specimens.

Since a coin does not allow the classic discovery identification in the form of an inscription without covering parts of the minting, based on around 1,000 optical features the software behind O.S.C.A.R. calculates an individual digital identification key that is as unique as a fingerprint for each coin recorded. This means that once a coin has been digitized, it can always be recognized again.

The digital recording of the find coin is based on photometric stereo analysis. It enables a reconstruction of the depth information of the object view. This method uses a series of shots of a fixed view of the object illuminated from different directions. The light dome is closed to protect the recordings from disturbing ambient light. The glass object support surface can be illuminated from below. This makes it easy to determine the contour of the coin and thus automatically cut it out.

O.S.C.A.R. consists of a high-resolution camera mounted at the zenith of a dome of 24 LED lights. This camera is height adjustable for focusing. A second camera is located under the object support surface for the coins. It is intended for determining the orientation of coins that have a circular contour. The weight of the find coin is determined using a precision scale and is read automatically.

A calibration body with colour wedges ensures the adjustment of the RGB colour values. Reference holes serve as scale representations. Several black ceramic balls are used to determine the positions of the LEDs in relation to the camera via reflections.

For digitization, the front and back of the find coin are placed on the support surface and recorded: Both sides can therefore always be assigned to a coin.

The centrally positioned main camera takes multiple images in an unchanged position, while the coin is illuminated from a different direction for each shot. A series of images with different exposure intensities are created for each of the LED lights in the light dome.

At the end, these image series are combined into HDR images (High Dynamic Range) in order to keep the final data volume as small as possible.

The images created in this way are the basis for the analysis process. Each pixel represents a small area element of the coin's surface. By evaluating the different lighting conditions, the inclination of this surface element can be estimated - the surface normal.

Non-reflective surfaces appear brightest when illuminated vertically by the light source. In addition, differences in brightness due to colour can be separated from those due to inclination - the albedo.

A height profile of the coin view can be reconstructed from the estimates for the surface normals. The edge of the coin cannot be captured with this setup.

The software behind O.S.C.A.R. calculates a unique, synthetic gray value image for each coin. The individual arrangement of the pixels in this image, which results from the minting characteristics of a coin and the individual damage, forms the digital fingerprint.

In order to find out during an inspection whether it is one and the same coin, the gray value images must be analyzed. Recognition takes place in several steps and initially by comparing the two coin contours: By comparing simple scalar features, such as diameter, deviations from circular shape, cracks and breakouts, the number of similar coins can be significantly limited.

In many ancient and medieval coins the contour is unique and easily distinguishable from that of other coins. With modern, machine-minted coins, differences in contours between coins of the same type are hardly measurable.

By precisely analyzing the two gray value images, an exact comparison can be made to identify a coin. The features of the two coin images are compared in a multi-stage process.

To do this, similar to when creating panoramic images, corresponding feature points in the images are searched for and compared. It starts with a low image resolution and is gradually increased.

If a sufficient number of feature pairs have been found, it is checked whether these points can be made to coincide even after rotating the gray value images. If this process fails, it is assumed that the coin is not the same.

However, if the highest resolution level is reached, the proportion of point pairs from the feature comparison that could be brought to coincide even after rotation is considered a measure of similarity.

The entire digitization process of a coin, including processing and securing the data, is completed in around 5 minutes - depending on the size of the coin. The finished images are then saved in TIFF format – the standard for long-term archiving – in publishable quality. Our post-process software processes the data and saves it in a specially developed database.

The images created can be looked at and edited in a special viewer. In order to simulate different lighting situations for coin analysis, the position of the light source above the coin surface can be controlled virtually in addition to the intensity.

Similar to turning a coin in sunlight or using the side-light method, the structure and embossed features of its surface can be viewed in detail - a good way to determine the type of find coins that are often corroded. In addition, weight, diameter and orientation, which were automatically recorded during digitization, can be displayed.