MultiMedia Database Lab

Ilaria Bartolini, Marco Patella, Karim Abdelcadir, Paolo Ciaccia

Application of real-time cinematographic replication in which the audience becomes the main character of the most suitable scenes.

SHIATSU is a framework software created for the automatic management of big collections of heterogeneous videos, able to supply effective and efficient tools for (i) automatic video segmentation in significant scenes and of every scene in representative key frames, employing both the visual information and the audio track of a video; (ii) the semantic automatic annotation of key frame, scenes and videos through simple keywords (tags) or more complex semantic concepts, such as taxonomy path (semantic tags); (iii) search for videos and/or scenes of interest through both keywords and "visual similarity" (e.g., "scenes where a particular actor is present - represented by one of his photos or by a key frame of a scene where he is present". The SHIATSU framework have many applications and it is used in various real contexts, such as the cinematographic one (there is a collaboration with the Cineteca of Bologna regarding that), the advertising one and the health and aging one.