Titelaufnahme

Titel
Computational analysis of petroglyphs / by Markus Seidl
VerfasserSeidl, Markus
Begutachter / BegutachterinBreiteneder, Christian
ErschienenWien, 2016
Umfangx, 212 Seiten : Illustrationen, Diagramme
HochschulschriftTechnische Universität Wien, Univ., Dissertation, 2016
Anmerkung
Zusammenfassung in deutscher Sprache
SpracheEnglisch
DokumenttypDissertation
Schlagwörter (EN)Computer Vision / Segmentation / Pixel Classification / Shape Descriptors / Surface Classification / Surface Description / Petroglyphs / Computing and Cultural Heritage / Valcamonica
URNurn:nbn:at:at-ubtuw:1-7852 Persistent Identifier (URN)
Zugriffsbeschränkung
 Das Werk ist frei verfügbar
Dateien
Computational analysis of petroglyphs [10.99 mb]
Links
Nachweis
Klassifikation
Zusammenfassung (Englisch)

Numerous petroglyphs have been pecked, scratched and carved into rock surfaces in the northern Italian valley Valcamonica. The classic documentation work carried out by archaeologists is a massively time-consuming process. The rising availability of digital images and 3D scans of petroglyphs facilitates digital workflows which can improve the documentation process. In this thesis, we aim at supporting the classic documentation pipeline for petroglyphs. The first step of the pipeline is the determination of the boundaries and spatial locations of petroglyphs on a rock surface. This is usually done by time-consuming manual contact tracing. Then, the found figures are classified according to their shapes and pecking styles. The large number of petroglyphs (Valcamonica contains up to 300.000 figures) demands large efforts for manual classification. The investigation of pecking styles is often impossible based on the contact tracings and thus requires researchers to return to the rocks frequently. Following the classic documentation pipeline for petroglyphs, we propose and evaluate novel methods. To determine the positions and shapes of petroglyphs on a rock panel we approach segmentation of 2D and 3D petroglyph images in pecked regions and natural rock surface. Furthermore, we use 3D scans to investigate the similarity of pecking styles, i.e. the shape, size, depth and spatial distribution of the peck marks a figure consists of. Finally, we develop a petroglyph shape descriptor which allows the classification of petroglyphs. Our tasks are challenging. The figures have been pecked over thousands of years. The rocks are subject to weathering and abrasion. Therefore, the visual and tactile appearance of the petroglyphs varies greatly. Figures have often been superimposed over existing figures. Consequently, many merged and partial figures exist. Contrary to previous work by others, we show that the segmentation of 2D images of rock surfaces is feasible. The employment of illumination-independent high-resolution 3D data of the surfaces- microtopographies clearly improves results. We facilitate the investigation of pecking styles by modeling their similarity with 3D surface descriptors. The shape classification of a dataset containing more than thousand petroglyphs yields very good results with a combination of skeleton-, boundary-, and regionbased shape descriptors. Our results can be useful for rock art researchers. Furthermore, we suggest how they can be applied in other domains.