Melanie Gau’s academic interests are determined by language (philology and work with “text” in the broadest sense), computer vision, as well as science and technology transfer

For many years she has worked at the interface between the humanities/arts and computer sciences, especially in the fields of computer linguistics, digital humanities, archaeology, multi-spectral imaging, digital image enhancement, and palaeoslavic studies. She took part in the international research projects The Sinaitic Glagolitic Sacramentary Fragments 2007–10 and its follow-up The Enigma of the Sinaitic Glagolitic Tradition 2011-14, and Textual Database Portal Project Manuskript (V.A. Baranov): Electronic Edition of the Psalterium Demetrii Sinaitici 2009-12

Investigation of Historical Manuscripts Based on Multi-Spectral Imaging

This work is concerned with the digitization and automated analysis of historical documents that are mainly in a poor condition. The investigation of these manuscripts is part of a series of interdisciplinary projects, involving philologist, chemists, and computer scientists. The degradations involve mould infection, background clutter, or water damage amongst others. Additionally, the manuscripts contain partially palimpsests (i.e. sections with erased and rewritten texts). Such degradations impede a transcription by philologist and worsen the performance of automated document image analysis techniques

In order to increase the visibility of the damaged writings, the documents are imaged with a portable Multi-Spectral Imaging (MSI) system. By applying MSI, the contrast of the faded out writings can be increased, compared to ordinary tungsten illumination. Post-processing techniques, such as dimension reduction tools, can be used to gain a further legibility increase and also to make the underwritten palimpsest text visible again or to differentiate between various ink types

We recently proposed a document binarisation method that makes use of the multi-spectral information in order to distinguish between different inks and the background. By incorporating spectral information, the method achieves an increased binarisation performance compared to state-of-the-art techniques, which are designed for grayscale or colour images. The multispectral images or resulting images of the post-processing methods are used as a basis for further document image analysis methods. Our project team has suggested several such techniques, involving Optical Character Recognition and writer identification. These methods are especially designed for historical documents in our projects

The current work gives an overview of the methods and findings gathered in our interdisciplinary projects at Computer Vision Lab (CVL) / TU Wien