Project Kenmerken Leren
MediaMill Semantic Video Search Engine
Video is the communication medium of the future, this is proven by sites like YouTube. But with the increasing number of videos available, it becomes more difficult to find them. To keep all videos accessible to everybody, access based on semantic cues has become a necessity. MultimediaN researchers have succeeded in developing the software for this and built the MediaMill Semantic Video Search Engine.
The Golden Demo shows, in an interactive way, how the search engine enables a user to explore large video archives quickly and with high precision.
While existing systems for video search mainly use subtitles, speech recognition or social tagging, the MediaMill Semantic Video Search Engine offers video access based on semantic content. This is accomplished by a self-learned lexicon of 500 visual concept detectors and an advanced video browser. In terms of quality and quantity, the detectors learned are state of the art. The system achieved a top ranking at the international TRECVID video competitions of 2004, 2005 and 2006. Also, in 2005 it received the Best Technical Demo Award at the prestigious ACM Multimedia Conference in Singapore.
The starting point for learning concepts is that an author creates a video with a certain intention. The intention of the video maker guides the production in all of its facets, resulting in a digital video. MultimediaN turns the authoring process around, paying attention to concepts like content, production style and context.
However, search behavior remains a complex issue. The search results should always be further explored interactively. To do so, the user is supported by advanced visualizations that help making mass video information more accessible.
In principle, the search engine can be used by everyone owning video material. For example, users with professional video archives (Beeld en Geluid), personal video archives, broadcasters, webcasters, and narrowcasters. Ilse Media and Fabchannel aim to deploy the semantic search engine to make video-files on internet and video-recordings of pop concerts easily accessible.
The same technique can be used to browse through music files. Aside from the MediaMill Semantic Video Search Engine, MultimediaN developed a Music Search Engine, in cooperation with the Meertens Institute. This musical search machine enables music lovers to find music by singing, playing, humming or whistling the melody, recognized by the computer
The technique can also be applied in automatic object recognition by intelligent cameras. In the Golden Video Object recognition by a Robot dog, a camera disguised as a dog learns to recognize objects that, for example, enable him to spot abandoned suitcases.
The Emotional Analyzer (N1)
Video Search Engine (N1)
Affective Mirror (N2)
Excercise in Immersion 4 (N2)
StreetTivo (N3 and N5)
The Investigator's Dashboard (N6)
The Surveillance Dashboard (N6)
On The Move (N6 and N9MI)
Concert Video Browser (N7 and N9MI)
Cultural Search Engine (N9C)