When you search information on the web by googling you use keywords. Keywords are found in text. No matter whether this is a blog, news article, or tags for photos. Viewdle helps you to find information in video. We know exactly who (person) and when (time with seconds precision) has appered in actual video frame. Thus you can effectively find information by finding video pieces with a person on screen, - a person who you associate with information you search.
Type a name of a person who you would like to find in video. We imply that you have one specific person in mind, therefore if several persons have names matching your query we will ask you which particular person did you mean. After unique person is identified search results will return a list of all videos (available in our system) which have searched person on-screen. Click on any title will call video player and will position playback to the moment of the first person appearance on screen in this video.
Indexed video files can then be searched using person names as queries. To try this out, just type a name (or part of it) in the search bar above. Try something like "Jack" or "David".
Top Charts shows people who where on screen most of the time (more than other people). Clicking on any of the person from top chart will lead you to searching this person in the same way as if you just typed the name of the person in a search bar.
You can narrow down your top charts by selecting certian channel or time period. For example, you want to know who where shown on screen on CNN channel during the hot pre-election month of September (Hillary Clinton, Barack Obama, etc) you have to select CNN in channels drop down list and set timeframe from 1st September 2007 to 30th September 2007.
Viewdle technology provides a new approach to searching video content through content-based identification of persons appearing in video. The Viewdle search engine is an automatic indexing process that allows video content to be searched using person names as queries. Current video search engines rely heavily on metadata from surrounding textual labels designated by original content owners. Thus, the results retrieved provide a text-based identification, rather than true visual content identification. Viewdle technology employs facial recognition algorithms to identify images within video content to ultimately expand the scope and accuracy of query results.
The facial recognition algorithms of the Viewdle indexing engine allow automatic visual analysis of video files that, in essence, identify persons in every frame of video. The engine is capable to scale to thousands of video (television) streams processed simultaneously and in real-time by looking through hundreds of thousands of traceable persons. The quality of the indexing engine is enhanced by Viewdle's ability to fuse several important technologies, including face recognition and context analysis (transcripts, closed captions, on-screen text, EPG), to optimize the accuracy and scope of results of persons appearing in video streams.
There are some limitations of the technology:
|
|