{"id":{"repo_id":"cau-kiel","oai_identifier":"oai:macau.uni-kiel.de:macau_mods_00008803"},"canonical_url":"https://search.dev.ndltd.org/etd/cau-kiel/oai:macau.uni-kiel.de:macau_mods_00008803","repository":{"repo_id":"cau-kiel","name":"Christian-Albrechts Universität Kiel","base_url":"https://macau.uni-kiel.de/servlets/OAIDataProvider"},"display":{"title":"From Bones to Bytes","abstract":"The digital acquisition of animal bone remains opens up new possibilities for reproducible computational analyses in archaeozoology. This thesis combines computer science, morphometrics, and archaeozoology in a unified framework spanning tabular osteometric data, landmark-based morphology, and landmark-free three-dimensional shape analysis. Using sheep and gazelle bones from modern and archaeological populations, it investigates sex-related variation, automated landmarking, and inter-species discrimination under conditions typical of archaeological material. The first part analyses osteometric measurements using machine learning methods to classify and explore sex-related structure in bone data. Supervised approaches generally performed well, with support vector machines proving the most robust under small-sample conditions, whereas unsupervised methods were primarily useful for exploratory analysis. The second part focuses on automated landmarking. Heuristic approaches based on image and mesh processing proved insufficient, whereas deep learning methods were markedly more effective. Mask R-CNN achieved good localisation accuracy on two-dimensional images, and an adapted multi-view landmark learning pipeline for 3D bone meshes produced promising results on unseen specimens, approaching or even surpassing human repeatability despite limited training data. The third part evaluates landmark-free approaches to three-dimensional shape comparison and classification. Several representations captured meaningful species-level differences between sheep and gazelle tali, while the newly proposed signed distance field descriptor (SDFD) emerged as a strong balance between performance, runtime, and interpretability. In summary, this thesis demonstrates that computational methods can meaningfully augment archaeozoological analysis across multiple levels of digital representation. Rather than replacing domain expertise, they serve as reproducible and scalable support tools that help formalise, accelerate, and extend traditional morphological research, thereby contributing to the development of morphoinformatics as an interdisciplinary framework for the digital analysis of animal bone remains.","abstract_html":"The digital acquisition of animal bone remains opens up new possibilities for reproducible computational analyses in archaeozoology. This thesis combines computer science, morphometrics, and archaeozoology in a unified framework spanning tabular osteometric data, landmark-based morphology, and landmark-free three-dimensional shape analysis. Using sheep and gazelle bones from modern and archaeological populations, it investigates sex-related variation, automated landmarking, and inter-species discrimination under conditions typical of archaeological material. The first part analyses osteometric measurements using machine learning methods to classify and explore sex-related structure in bone data. Supervised approaches generally performed well, with support vector machines proving the most robust under small-sample conditions, whereas unsupervised methods were primarily useful for exploratory analysis. The second part focuses on automated landmarking. Heuristic approaches based on image and mesh processing proved insufficient, whereas deep learning methods were markedly more effective. Mask R-CNN achieved good localisation accuracy on two-dimensional images, and an adapted multi-view landmark learning pipeline for 3D bone meshes produced promising results on unseen specimens, approaching or even surpassing human repeatability despite limited training data. The third part evaluates landmark-free approaches to three-dimensional shape comparison and classification. Several representations captured meaningful species-level differences between sheep and gazelle tali, while the newly proposed signed distance field descriptor (SDFD) emerged as a strong balance between performance, runtime, and interpretability. In summary, this thesis demonstrates that computational methods can meaningfully augment archaeozoological analysis across multiple levels of digital representation. Rather than replacing domain expertise, they serve as reproducible and scalable support tools that help formalise, accelerate, and extend traditional morphological research, thereby contributing to the development of morphoinformatics as an interdisciplinary framework for the digital analysis of animal bone remains.","abstract_has_math":false,"creators":["Schüler, Nadine Sarah"],"institution":"Christian-Albrechts-Universität zu Kiel","degree_name":null,"degree_level":"thesis.doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":["Kröger, Peer","Peters, Joris","Renz, Matthias"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-07-08","date_published":"2026-07-08","updated_at":"2026-07-24T01:35:26Z","subjects":["Morphoinformatics","Morphometrics","Data Science","Supervised Learning","Unsupervised Learning","Deep Learning","Animal Bone Analysis","Machine Learning","AI","Artificial Intelligence","Landmarking","Shape Descriptor"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://macau.uni-kiel.de/receive/macau_mods_00008803","outbound_label":"Repository record","outbound_source":"source_url"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Kröger, Peer","Peters, Joris","Renz, Matthias"]},{"key":"dc:creator","label":"Author","values":["Schüler, Nadine Sarah"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:publisher","label":"Institution","values":["Universitätsbibliothek Kiel"]},{"key":"dc:type","label":"Dc Type","values":["PhDThesis"]},{"key":"thesis:degree_level","label":"Degree Level","values":["thesis.doctoral"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Christian-Albrechts-Universität zu Kiel"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Morphoinformatics","Morphometrics","Data Science","Supervised Learning","Unsupervised Learning","Deep Learning","Animal Bone Analysis","Machine Learning","AI","Artificial Intelligence","Landmarking","Shape Descriptor"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["The digital acquisition of animal bone remains opens up new possibilities for reproducible computational analyses in archaeozoology. This thesis combines computer science, morphometrics, and archaeozoology in a unified framework spanning tabular osteometric data, landmark-based morphology, and landmark-free three-dimensional shape analysis. Using sheep and gazelle bones from modern and archaeological populations, it investigates sex-related variation, automated landmarking, and inter-species discrimination under conditions typical of archaeological material. The first part analyses osteometric measurements using machine learning methods to classify and explore sex-related structure in bone data. Supervised approaches generally performed well, with support vector machines proving the most robust under small-sample conditions, whereas unsupervised methods were primarily useful for exploratory analysis. The second part focuses on automated landmarking. Heuristic approaches based on image and mesh processing proved insufficient, whereas deep learning methods were markedly more effective. Mask R-CNN achieved good localisation accuracy on two-dimensional images, and an adapted multi-view landmark learning pipeline for 3D bone meshes produced promising results on unseen specimens, approaching or even surpassing human repeatability despite limited training data. The third part evaluates landmark-free approaches to three-dimensional shape comparison and classification. Several representations captured meaningful species-level differences between sheep and gazelle tali, while the newly proposed signed distance field descriptor (SDFD) emerged as a strong balance between performance, runtime, and interpretability. In summary, this thesis demonstrates that computational methods can meaningfully augment archaeozoological analysis across multiple levels of digital representation. Rather than replacing domain expertise, they serve as reproducible and scalable support tools that help formalise, accelerate, and extend traditional morphological research, thereby contributing to the development of morphoinformatics as an interdisciplinary framework for the digital analysis of animal bone remains.","Die digitale Erfassung tierischer Knochenreste eröffnet neue Möglichkeiten für reproduzierbare, computergestützte Analysen in der Archäozoologie. Im Zentrum steht dabei die Frage, wie künstliche Intelligenz archäozoologische Forschung auf unterschiedlichen Ebenen digitaler Repräsentation unterstützen kann. Die Arbeit verbindet Informatik, Morphometrie und Archäozoologie in einem gemeinsamen methodischen Rahmen, der tabellarische osteometrische Daten, landmarkenbasierte Morphologie und landmarkenfreie dreidimensionale Formanalyse umfasst. Der erste Teil widmet sich der Analyse osteometrischer Messdaten mit Methoden des maschinellen Lernens, um geschlechtsbezogene Strukturen in den Knochendaten zu identifizieren. Überwachte Lernverfahren erzielten insgesamt gute Ergebnisse, wobei sich Support-Vector-Maschinen insbesondere bei kleinen Stichproben als besonders robust erwiesen. Unüberwachte Verfahren waren vor allem für explorative Analysen hilfreich. Im zweiten Teil steht die automatisierte Landmarkensetzung im Mittelpunkt. Während heuristische Ansätze keine zufriedenstellenden Ergebnisse lieferten, erwiesen sich Deep-Learning-Methoden als deutlich erfolgreicher. Mask R-CNN erzielte auf zweidimensionalen Bildern eine gute Lokalisierungsgenauigkeit, und eine angepasste Multi-View-Landmark-Learning-Pipeline für dreidimensionale Knochenscans lieferte auch bei begrenzten Trainingsdaten vielversprechende Ergebnisse. Der dritte Teil untersucht landmarkenfreie Verfahren zum dreidimensionalen Formvergleich und zur Klassifikation. Dabei konnten mehrere Repräsentationen artspezifische Unterschiede zwischen Schaf- und Gazellentali erfassen. Besonders hervorzuheben ist der neu entwickelte Signed Distance Field Descriptor (SDFD), der eine Balance zwischen Leistungsfähigkeit, Laufzeit und Interpretierbarkeit bietet. Zusammenfassend zeigt die Dissertation, dass computergestützte Methoden archäozoologische Analysen sinnvoll ergänzen können. Sie ersetzen fachliche Expertise nicht, sondern erweitern sie durch reproduzierbare und skalierbare Werkzeuge und leisten damit einen Beitrag zur Entwicklung der Morphoinformatik als Rahmen für die digitale Analyse tierischer Knochenreste."]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["From Bones to Bytes"]}]}],"canonical_facts":{"dc:contributor":["Kröger, Peer","Peters, Joris","Renz, Matthias"],"dc:creator":["Schüler, Nadine Sarah"],"dc:description.abstract":["The digital acquisition of animal bone remains opens up new possibilities for reproducible computational analyses in archaeozoology. This thesis combines computer science, morphometrics, and archaeozoology in a unified framework spanning tabular osteometric data, landmark-based morphology, and landmark-free three-dimensional shape analysis. Using sheep and gazelle bones from modern and archaeological populations, it investigates sex-related variation, automated landmarking, and inter-species discrimination under conditions typical of archaeological material. The first part analyses osteometric measurements using machine learning methods to classify and explore sex-related structure in bone data. Supervised approaches generally performed well, with support vector machines proving the most robust under small-sample conditions, whereas unsupervised methods were primarily useful for exploratory analysis. The second part focuses on automated landmarking. Heuristic approaches based on image and mesh processing proved insufficient, whereas deep learning methods were markedly more effective. Mask R-CNN achieved good localisation accuracy on two-dimensional images, and an adapted multi-view landmark learning pipeline for 3D bone meshes produced promising results on unseen specimens, approaching or even surpassing human repeatability despite limited training data. The third part evaluates landmark-free approaches to three-dimensional shape comparison and classification. Several representations captured meaningful species-level differences between sheep and gazelle tali, while the newly proposed signed distance field descriptor (SDFD) emerged as a strong balance between performance, runtime, and interpretability. In summary, this thesis demonstrates that computational methods can meaningfully augment archaeozoological analysis across multiple levels of digital representation. Rather than replacing domain expertise, they serve as reproducible and scalable support tools that help formalise, accelerate, and extend traditional morphological research, thereby contributing to the development of morphoinformatics as an interdisciplinary framework for the digital analysis of animal bone remains.","Die digitale Erfassung tierischer Knochenreste eröffnet neue Möglichkeiten für reproduzierbare, computergestützte Analysen in der Archäozoologie. Im Zentrum steht dabei die Frage, wie künstliche Intelligenz archäozoologische Forschung auf unterschiedlichen Ebenen digitaler Repräsentation unterstützen kann. Die Arbeit verbindet Informatik, Morphometrie und Archäozoologie in einem gemeinsamen methodischen Rahmen, der tabellarische osteometrische Daten, landmarkenbasierte Morphologie und landmarkenfreie dreidimensionale Formanalyse umfasst. Der erste Teil widmet sich der Analyse osteometrischer Messdaten mit Methoden des maschinellen Lernens, um geschlechtsbezogene Strukturen in den Knochendaten zu identifizieren. Überwachte Lernverfahren erzielten insgesamt gute Ergebnisse, wobei sich Support-Vector-Maschinen insbesondere bei kleinen Stichproben als besonders robust erwiesen. Unüberwachte Verfahren waren vor allem für explorative Analysen hilfreich. Im zweiten Teil steht die automatisierte Landmarkensetzung im Mittelpunkt. Während heuristische Ansätze keine zufriedenstellenden Ergebnisse lieferten, erwiesen sich Deep-Learning-Methoden als deutlich erfolgreicher. Mask R-CNN erzielte auf zweidimensionalen Bildern eine gute Lokalisierungsgenauigkeit, und eine angepasste Multi-View-Landmark-Learning-Pipeline für dreidimensionale Knochenscans lieferte auch bei begrenzten Trainingsdaten vielversprechende Ergebnisse. Der dritte Teil untersucht landmarkenfreie Verfahren zum dreidimensionalen Formvergleich und zur Klassifikation. Dabei konnten mehrere Repräsentationen artspezifische Unterschiede zwischen Schaf- und Gazellentali erfassen. Besonders hervorzuheben ist der neu entwickelte Signed Distance Field Descriptor (SDFD), der eine Balance zwischen Leistungsfähigkeit, Laufzeit und Interpretierbarkeit bietet. Zusammenfassend zeigt die Dissertation, dass computergestützte Methoden archäozoologische Analysen sinnvoll ergänzen können. Sie ersetzen fachliche Expertise nicht, sondern erweitern sie durch reproduzierbare und skalierbare Werkzeuge und leisten damit einen Beitrag zur Entwicklung der Morphoinformatik als Rahmen für die digitale Analyse tierischer Knochenreste."],"dc:format.medium":["application/pdf"],"dc:publisher":["Universitätsbibliothek Kiel"],"dc:subject":["Morphoinformatics","Morphometrics","Data Science","Supervised Learning","Unsupervised Learning","Deep Learning","Animal Bone Analysis","Machine Learning","AI","Artificial Intelligence","Landmarking","Shape Descriptor"],"dc:title":["From Bones to Bytes"],"dc:type":["PhDThesis"],"thesis:degree_level":["thesis.doctoral"],"thesis:institution_name":["Christian-Albrechts-Universität zu Kiel"]},"updated_at":"2026-07-24T01:35:26Z"}