Back to search

National Technical University of Athens (NTUA)

Ψηφιακή επεξεργασία εικόνας για τη σχεδιαστική ταυτοποίηση σχολών ιστορικών νωπογραφικών

Abstract

dc:description

The objective of this current research is the classification, recognition and identification of historical frescoes of the 13th and 14th centuries from the most important urban centers of Byzantium through digital image processing. The subject is interdisciplinary. It is examined from Byzantine art and its distinguishing features, as well as from technologies involved in the identification and methodology, specifically in digital image processing with validation in colorimetric analysis, as formulated by N. Vesic’s doctoral thesis. The doctoral thesis was created as an additional tool for art scholars on the identification and analysis of works of art by using the appropriate tools, as detailed here. In this context, the steps followed include: • the creation of the database, which consists of both photographic material of frescoes and information for each image. The photographic material are frescoes from the most important monuments of the Palaeologian period and some images of frescoes from the post-Byzantine period of the Cretan School have also been included. 123 color images of frescoes were collected • creating an application in Python to manage and analyze the data. Particular emphasis was placed on the accuracy and verification of the information entered into the database to ensure the reliability of the results. More specifically the elements are organized into three main entities, Picture, Work, Artist. This leads to the conclusion that the information material in the database is the source of truth for each image and is an important success factor of the machine learning model. • it was considered of utmost importance to enrich the database by dividing the images into smaller patches, with the result that the final images total 13252. Each of the smaller images has the corresponding informational material. In this way the database was enriched and grew. • After that, AlexNet, a Convolutional Neural Networks (CNN), was used for accurate image categorization, the Adam optimization algorithm as well as the F1 metric. The system was trained on all data in the learning set and produced an identification rate of 85%. From the results of the plots, the model behaves satisfactorily and overfitting was avoided • spikes in the validation score F1 curve indicate some variability in model performance across the validation set that can be addressed using techniques such as cross-validation, robust data augmentation, or normalization methods. • the realization of the colorimetric analysis of the images from the database in two pilot applications: a) images from various frescoes, b) images of frescoes by well-known artists of Byzantine art, Manuel Panselinos and Michael Astrapas and Eutychios in order to achieve a colorimetric approach with the images of frescoes by an unknown artist, in the chapel of Agios Euthymios whose frescoes are also from the Macedonian School The results in both pilot applications show a very good correlation of the parameters Δb* and ΔC*, especially between the frescoes of the Unknown artist and Manuel Panselinos. Specifically, the overall color parameter of the images from the artists' frescoes (ΔC*) correlates more with the shade of yellow (Δb*) but with a different intensity. This intensity is mathematically related to the slope of the straight correlation line in the diagram Δb* ΔC* which for Manuel Panselinos is 1.15 but for Michael Astrapas and Euthychios it is 0.78.In the second application of the colorimetric analysis, 14 fresco images of the Macedonian School, from the chapel of Agios Euthymios in Thessaloniki, were examined. The specific temple was chosen, on the one hand, because it is one of the most interesting ensembles of the Paleologian renaissance, and on the other hand, there are studies that support that the frescoes were made by Manuel Panselinos. The studies are based on the similarity, both stylistic and rendering, of the elements with the frescoes from the temple of Protatus. From the application of the colorimetric method, it is concluded that the images of the frescoes used have common color characteristics and do not deviate from each other. This finding confirms the characterization of the Schools, as the Byzantine frescoes examined bear common elements. In this way, the macroscopic observations of art historians regarding the analysis of the color palette of the artists during Byzantine times and the distinction of artistic Schools are confirmed. This doctoral thesis is a comprehensive proposal on digital image analysis and has potential for development, mainly by enriching the database with both photographic material and additional informational elements.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Vlysidi, Antigoni
  • Βλυσίδη, Αντιγόνη

Subjects

dc:subject × 24

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/57606

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Vlysidi, Antigoni; Βλυσίδη, Αντιγόνη. Ψηφιακή επεξεργασία εικόνας για τη σχεδιαστική ταυτοποίηση σχολών ιστορικών νωπογραφικών. National Technical University of Athens (NTUA), 2024. http://hdl.handle.net/10442/hedi/57606