2020 IASA - FIAT/IFTA Joint Conference

The German Broadcasting Archive (Deutsches Rundfunkarchiv, DRA) preserves, among other things, the archives of GDR television from the years 1952 to 1991. The image quality of the historical material is very heterogeneous - the collections include both colour and black-and-white material, low-resolution images and, in some cases, even screen recordings. This poses challenges for video mining software.

In the DFG-funded research project „Visual Information Retrieval in Video Archives (VIVA)“, the DRA, the Philipps University of Marburg and Leibniz Information Centre for Science and Technology (TIB) in Hannover take on this challenge. Currently 100 concepts and 100 persons are being trained to be recognized in the material of GDR television. A similarity search (query by image) is in test operation.

As part of the project, a software is developed that integrates all steps from the acquisition of training material, through the annotation of individual images, to training and management of the models. The goal is to enable archivists to carry out the entire training process on a user-friendly interface and to be able to independently add new people or concepts to the training when new requirements arise. In contrast to generic mining offers such as Microsoft Indexer, archivists can determine all steps themselves.

The software will be published under an open source license at the end of the project in 12/2020 and will be available to all interested parties. The treasures lying hidden in archives around the world have become much more accessible thanks to digitisation projects – in this context the use of this software for archives has great potential.