Finding Images and Searching Videos
by
Randy Mears
When I search the Internet for text, using a tool like Google, I am amazed by how much pertinent information I find. Google is a great tool when coupled with the skills required to devise accurate and effective search arguments. There are a number of interesting operators that can make the formulation of a search argument even more interesting. Google’s on-line cheat sheet has a number of operators that really help narrow things. Sure, text searching will continue to improve, but it is already good enough for most of us.
As the Internet’s inventory of images increase, categorization and keyword extraction strategies will need to be devised to make these images findable. Google is already labeling images using human intelligence as a resource. Based on searches that I have launched on the Google Images Web site, I am satisfied that text based search arguments work fairly well in locating specific images. A search for “london bridge at sunset,” sans quotes, returns this multi-page list. The success of such a search is due in large part to keywords and other forms of metadata found in context with the image. Images without context are a more complex problem but using human intelligence to define keywords seems to help fill that gap.
Video poses a more complex problem because simply finding it may not be enough. You may be searching for a specific scene or segment, something that is just a small part of a large video. A researcher at the University of Leeds has presented research that approaches the problem with a combination of script data, close-caption text and sophisticated face tracking software. It's only the beginning. Now, with Google and YouTube together, the future for video searching looks brighter than ever.
