Techn. Fakultät Willkommen am Institut für Informatik FAU-Logo

Daniel Stromer M. Sc.

Researcher in the Image Analysis (IMA) group at the Pattern Recognition Lab of the Friedrich-Alexander-Universität Erlangen-Nürnberg

3-D Reconstruction of Historical Documents

The state-of-the-art digitization imaging process for documents is done by a scan robot with a high resolution camera system where the pages are automatically turned by an air-flow. This technique requires that page-turning of the document is easily possible, which is not applicable for a lot of historical documents. Aging processes and damage caused by external influences make the digitization of some historical documents considerably more difficult. For this hard cases, non-invasive 3-D imaging approaches can help to reveal the hidden contents.

Historical ink recipes differ from modern ones as dyes were not as widely accessible as they are nowadays. Many historical recipes are based on metallic ingredients such as iron (iron gall ink) or copper (malachite) and also the water used to mix inks was polluted by particles such as lead (see research on Herculaneum inks). Those metallic ingredients allow 3-D X-ray scanners to differentiate it from the writing media (parchment, paper, papyrus) as the attenuation is rather high compared to cellulose or collagen. 

For our research, we use conventional 3-D X-ray CT systems for the reconstruction of document images which are more mobile than Synchrotron systems. Our first experiments, tested on a medical X-ray C-Arm CT robotic system, showed that the 3-D reconstruction of the ink was possible [Opens external link in new windowIVCNZ], but the imaging system was too inaccurate, such that we were only able to image a very small area of the ink. As it was not possible to set parameters to the range that we needed, we choose an X-ray system that is built for material testing [Opens external link in new windowICT]. At the moment, we are in the process of refining the scan parameters to find a trade-off between applied dose and readability of the information written or drawn on the paper.

To investigate the resulting output, the generated 3-D volume has to be processed such that the pages are flattened and mapped to 2-D. Therefore, a fully automatic algorithm has been developed separating and extracting the pages without the need of user interaction providing a 2-D mapped image for each page. The following Figure shows four exemplary pages that were written by our own. The page snippets have a width and height of approximately 4 cm with a letter height ranging from 2 cm to 0.25 cm. The book consists of 22 pages. The left image shows the original book, while the right images show the real page (black writings) and its automatically segmented and reconstructed CT result (bright writings). This means that we are able to read a the book's writings without the need to open it. Test with a larger book were already made and showed good results. The images will be updated when the results are published.




Segmentation of Fat and Fascia Layers in Ultrasound Images

The connective tissue between the fat layer and the skin termed fascia has been of interest to the clinical and zoological research to study normal skin echogenicity, thickness and hydration status, as well as the echogenicity patterns of various pathological conditions. The current state-of-the-art method for visualizing these layers is to use ultrasound imaging. By visual inspection of those (Figure below), one can see four different layers: skin, fat, fascia and muscle. Our goal is to separate the different layers fully automatically by applying appropriate segmentation algorithms. Furthermore, we want to provide a GUI for specialist such that there is no more need to manually measure the layers.