CNN-Based Page Segmentation and Object Classification for Counting Population in Ottoman Archival Documentation.

Area of Science:

• Computer Science
• Digital Humanities
• Archival Science

Background:

• Digitalization of historical archives necessitates robust analysis systems.
• Page segmentation and layout analysis are critical for Optical Character Recognition (OCR) and handwritten text recognition.
• Challenges include document degradation, digitization errors, varied layouts, and unique Arabic script properties.

Purpose of the Study:

• To develop an automatic system for counting registered individuals in historical documents.
• To assign counted individuals to specific populated places.
• To address the complexities of analyzing historical Arabic script documents.

Main Methods:

• A Convolutional Neural Network (CNN)-based architecture was developed.
• A labeled dataset was created from Ottoman Empire population registers (1840s-1860s).
• The system was trained to classify objects and count individuals.

Main Results:

• The system achieved promising results in object classification.
• Accurate counting of individuals was demonstrated.
• Successful assignment of individuals to populated places was achieved.

Conclusions:

• The developed CNN-based system effectively analyzes historical Ottoman registers.
• This approach offers a significant advancement for historical document analysis, particularly for Arabic scripts.
• The system demonstrates potential for improving demographic reconstruction from archival data.