Our comments and trackback policy You Link We Follow, You Comment We Promote
VideoSurf, a video search startup that unveiled today at TC 50, is attacking the video search problem in a new way. Instead of relying upon users to tag the content and than categorize the videos and make them relevant the service is digging into the videos themselves using computer vision.
This definitely is a better way in a sense that it does not squarely relies on users to do the tagging, but it would be as good as the algorithms themselves and we have yet to see how good they really are. This in no way is a negative comment on the startup since Google relies solely on algorithms to determine the relevance of a website in response to search queries.
VideoSurf actually meshes a mixture of things. It tries to recognize what’s in an image and than goes on further to break the video into frames and tries to classify each of the frames by topic. This in depth digging allows VideoSurf to recognize faces much more accurately than its competitors such as CastTV, BlinkX or Like.com.
As a search result the service displays a series of frames from the video in thumbnail format to the user, so that he could have a relatively good idea about what’s in the video just by viewing the summary. Not only does this aids in at a glance information retrieval, it also allows users to start a certain movie from a specific spot that he might be most interested in.
If they could somehow bundle it with language processing, we would also be able to search a movie based on what has been said in it. This would in turn create an ultimate video search machine, something that Google is most likely to create due to its huge assortment of videos – thanks to Youtube. The reason such a collection is important because machine learning techniques take a huge amount of data and a couple of years of iterations to perfect.
Lior Delgo, who previously founded FareChase, a travel search engine that Yahoo acquired in 2004 is behind this new venture as well. Given his experience we could be a little sure that VideoSurf might get to somewhere. With a near perfect video search index, VideoSurf would become the same thing for web video that Google is for web sites – a gateway. This in turn would diminish the importance of Youtube, because in order to find a video on Youtube you would have to come to VideoSurf.
VideoSurf like Google is a brainchild of some brilliant scientists, which includes Achi Brandt, who is behind the computational algorithms that churns out huge amounts of visual data in milliseconds and Eitan Sharon, who is behind the visual search techniques the search engine is employing.
In addition to this the company is financially backed by former Vice President Al Gore and has raised $5.5 million in a single round so far. The company is based in San Mateo, California.
You can watch their TC 50 launch video below:
