Mathias Kirsten

Movies are a special kind of data. Moving images and language meet. A movie contains a lot of information, but it is very compressed in the producer's description. For different use cases, however, a deeper scene-level information level may be necessary. In an exploratory project, we investigated how meta-data can be extracted directly from the footage using techniques such as computer vision and NLP. For this purpose, a multi-stage procedure was developed to first cuts films into scenes and then analyze the scene content. A semantic search is used to check the plausibility of the results and thus clarify the points at which AI and humans understand the same thing but mean something different. Limits and possibilities are shown and further development opportunities are highlighted.