Looking back to 1850 in 2025: Historascan to digitize historical journals

Abstract

This chapter analyses current technologies and the challenges involved in extracting and classifying articles and news headlines from historical journals, as well as converting images to text format. The work to develop a tool focused on digitising historical journals was carried out by a multidisciplinary team of experts in media studies, artificial intelligence, image processing, and cultural heritage preservation. The data used derives from two historic Portuguese journals, Diário de Notícias and Jornal de Notícias, which were created in the mid-19th century. This project is based on a mixture of heuristics, computer vision, pattern recognition, and other artificial intelligence and machine learning techniques. The main challenges included the variability in the design of historical journals, preserving the quality of images over time, and continuously improving image processing and OCR techniques to adapt to different styles and periods of newspapers.