This week my interview is with Mike Nichols of Microsoft. We spoke about image search, video search, and celebrity search. With Microsoft’s recent set of releases, their new video search engine was unveiled for the first time. One of the more interesting aspects of Microsoft’s video search engine is the smart preview feature.
Mike showed me a great demo of a Baron Davis dunk (then select the video titled “Baron Davis dunks on AK-47 in Game 3”). The video is 25 seconds long, but the dunk happens only at the end.
The smart trailer technology is able to algorithmically figure out that the focal point of the video is at the end. As a result, the trailer Microsoft offers at the video selection screen goes right to the most important part of the video, to enable the user to better decide is this is a video that they want to watch.
It really sparks some interesting thoughts about the nature of the algorithms to determine the most important part of a video. One possible approach is simply to evaluate the video as a whole and then use a set of algorithms to figure out what part of the video is the most different from the rest of the video.
Whatever techniques Microsoft decided to use, it’s clear that there are a large number of different scenarios that they have to deal with. In the case of the Baron Smith dunk, there is clearly a significant increase in crowd noise right at the end of the video. However, not every video provides a clue as simple as an great increase in crowd noise.