Historical Manuscripts Analysis: A Deep Learning System for Writer Identification Using Intelligent Feature Selection...

Area of Science:

• Computer Science
• Digital Humanities
• Historical Document Analysis

Background:

• Identifying historical manuscript writers is vital for historical research and solving mysteries.
• Existing methods may lack the precision needed for complex historical document analysis.

Purpose of the Study:

• To develop and evaluate a deep learning system for historical manuscript writer identification.
• To assess the effectiveness of vision transformers and feature selection techniques in this domain.

Main Methods:

• Document preprocessing included bilateral filtering and Otsu thresholding.
• Feature extraction used the FAST detector and k-means clustering for uniform patches.
• Vision transformer models were employed for classification of handwriting patterns.

Main Results:

• The system demonstrated robust performance in classifying historical manuscripts by writer.
• Vision transformers showed superior capability in learning complex patterns from manuscript data.
• The approach outperformed state-of-the-art methods on the ICDAR 2017 dataset.

Conclusions:

• The developed deep learning system is a powerful tool for historical manuscript analysis.
• Vision transformers represent a significant advancement in automated historical document analysis.
• This research offers a novel solution for historical writer identification challenges.