Supervisors: Florian Kleber, Marco Peer
Status: open
The analysis of artistic styles is a well-established challenge in the Cultural Heritage field. Traditional artworks, such as statues and paintings, have been thoroughly studied, leading to standardized methods for identifying an artist’s style. However, historical violins present a unique case. While their main stylistic features are known, only a limited number of experts can reliably attribute a violin to its maker, with occasional disagreements even among scholars.
Figure 1: Images for Stradivari “Clisbee” (1669): (a) front view; (b) soundboard; (c) back plate; (d) head, front side; (e) head, treble side; (f) head, back side. Image taken from [1]
In this thesis, a large dataset of violins from different makers, including images from various perspectives and features, such as scroll and body views as shown in Figure 1, is investigated. The aim is to implement a retrieval system that allows scholars to identify common features between a query violin and those in the database more efficiently. Since the dataset comprises violins from up to 1,000 makers, modern deep learning approaches should be employed, with a focus on identifying similarities in the violins’ properties. The thesis is in cooperation with the violin experts/makers Julia Pasch, Lüder Pasch, and Marcel Richters who are also providing the data.
[1] Dondi et al. “What do luthiers look at? An eye tracking study on the identification of meaningful areas in historical violins,” Multimedia Tools and Applications, 78, 19115–19139, 2019.
The thesis will include the following components:
- Dataset Review and Literature Survey
A comprehensive review of the violin dataset and related research. - Selection of a Classification and Retrieval Approach
Choosing an appropriate method for classification and retrieval, such as a full-image-based model that can later be adapted to focus on specific violin features. - Implementation and Evaluation
Developing and testing a system with a graphical interface to visualize the retrieval, potentially highlighting similarities between violins. - Written Report and Final Presentation
Completion of the written thesis.
Prerequisites/Helpful Experience
- Proficiency in Python
- Strong understanding of deep learning concepts
- Experience with machine learning frameworks (preferably PyTorch)