An Experimental Comparison between Deep Learning and Classical Machine Learning Approaches for Writer Identification...

Area of Science:

• Digital Humanities
• Computer Vision
• Palaeography

Background:

• Advancements in digital imaging and algorithms enhance palaeography tools.
• Feature selection is critical but challenging due to document variability.
• Deep learning (DL) offers automatic feature extraction for ancient document analysis.

Purpose of the Study:

• To evaluate DL as a general methodology for palaeography applications.
• To compare DL performance against classical machine learning.
• To test approaches on a challenging dataset of the 12th-century Avila Bible.

Main Methods:

• Comparative analysis of DL and classical machine learning.
• Application to scribe identification in manuscript images.
• Utilizing a large dataset from the Avila Bible.

Main Results:

• DL approaches demonstrated effectiveness in scribe identification.
• Comparison highlighted DL's ability to generalize without prior knowledge.
• The Avila Bible dataset served as a robust test case.

Conclusions:

• Deep learning presents a viable general methodology for palaeography.
• DL systems can automate feature extraction, simplifying analysis.
• Further research can refine DL applications for historical document studies.