It’s still in private beta, but my colleague Beti Cung and I have seen demos of Viewdle’s impressive facial-recognition-based video search technology. It’s a full search engine on all video feeds that use face recognition/speech2text/OCR in the ability to do in-video facial recognition. The company provides frame-by-frame indexing that can take a viewer directly to the video clip where a subject actually appears without relying on meta-tags or speech-to-text indexing. For example, their engine can search and show automatically a video of Hillary Clinton when she pronounces single words or phrases such as "President" or "elections", or "I don't get up every morning to go out and make a great speech". The search can be run on real-time feeds or against archived assets. The company’s IP is based on 35 years of research funded by the KGB with patents in preparation. The development team is based at the University of Ukraine in Kiev and the company’s corporate headquarters are in New York.

Viewdle’s engine indexes 55 frames per second (roughly equivalent to 1 viewable second of video processed per second) and can process a thousand streams simultaneously with a 1:1 stream to server ratio. On the conservative-side, the company should be able to process 24,000 hours of video in a day compared to 6000+ hours a day for a competing company that does speech-to-text video indexing. Viewdle is already working with Reuters lab, and has received recognition in the market: they were selected to present as a TechCrunch40 in September 2007 and were one of the finalists at the Crunchies 2007 in the category "Best Technology Innovation/Achievement". The Viewdle business model focuses on white-labeling their solution and monetization via search ad-revenue of video assets. Other companies in the video search space, such as Truveo, Blinkx, CastTV and EveryZing –do not employ facial recognition in their video search function. Companies in the facial recognition area include Polar Rose, Neven Vision (acquired by Google in 2006) and Riya (which has abandoned much of its face recognition in photos efforts) – none of whom currently provide in-video facial recognition.