Felix Saurbier, Prof. Dr. Ralph Ewerth
1 March 2017 – 31 December 2017
The technology currently used for Video Concept Detection (VCD) in the TIB AV portal is to be replaced by a new module being researched and developed at TIB.
In order to improve detection rates and to increase the number of recognized image concepts, the VCD module currently used in the AV portal is to be replaced by new technologies developed at TIB. The underlying method is based on trained neural networks and is developed by the "Visual Analytics" research group at TIB. For use in the AV portal, the system is first trained on 75 relevant image concepts from the various TIB domains (e. g. "experiment" or "technical drawing") and then integrated into the productive system. The new VCD module will be based on modern machine learning methods and deep neural networks, which are able to analyse and classify audio-visual material with great accuracy. The integration of these technologies will further improve the automatic indexing and searchability within scientific videos of the AV portal. In addition, the technologies used offer the possibility of continuously expanding the image concepts detected and of successively optimizing the respective recognition rates.