Archive users searching for a specific image or scene are often confronted with ten thousands of hours of footage. How can an archive provide its users with a possibility of effective retrieval? Today, the answer is: by massively investing in staff intellectually documenting the material. A colossal task, hardly manageable for smaller institutions with limited staff resources. The project VIVA – Visual information retrieval in video archives has a goal: to support archives in this task by developing an innovative videomining software. Using Artificial Intelligence (AI) the software automatically indexes digitized archival footage – thereby creating new opportunities for information retrieval concerning scenes, people and similar images in so far undocumented material.
Supported by Deutsche Forschungsgemeinschaft (DFG) three partners work on the project in close cooperation: the research team Visual Analytics at Leibniz Information Centre for Science and Technology (TIB), the Department for Distributed Computing at Marburg University and the German Broadcasting Archive (DRA). The researchers use modern deep learning architectures to ensure the effective training of Deep Convolutional Neural Networks. The DRA provides the project with valuable ground truth data and DRA’s archivists and documentarists bring their firsthand experience with the needs of journalists and scientists using the archive to the table.
The VIVA tool is available at https://github.com/TIBHannover/VIVA.
- Screenshot »Aktuelle Kamera: Politische und wirtschaftliche Lage in Westberlin« (1966), DRA (IDNR 331546)
- Berlin, Alexanderplatz, "Haus des Lehrers", Fernsehturm, Bundesarchiv, Bild 183-H1002-0001-002 / CC-BY-SA 3.0
- Erich Honecker, Bundesarchiv, Bild 183-R1220-401 / Unknown / CC-BY-SA 3.0
- FDJ Fahne, Vwpolonia75, CC BY 3.0
- Ortsschild Berlin, Monster4711, CC0 1.0