Google: Baby Steps with Image Recognition
August 31, 2009
With attention focused on Google Books and Google lobbying, modest technical innovations can be overlooked. The Overflight service flagged US7,580,568 “Methods and Systems for Identifying an Image as a Representative Image for an Article.” On the surface, what is the big deal about parsing a document with multiple images and taking one as a representative image? Google does this frequently. Navigate to Google News and look at the images positioned next to a news story.
Now what if an article has an image but that image is not one that represents the information in the article? In the good old days when traditional publishers were kings, a human would flip through a photo archive, locate a suitable image, and mark up the copy to show the compositor where to put the picture. Google has automated this service. (Page 12, Column B, line number 49.) Not a big deal, but it is one that chops costs out of the process of assembling original mash ups of information.
One of the principal findings from my research into Google’s technology is that the company has been purposeful in squeezing costs out of operations that are often money bottlenecks when traditional methods are shoehorned into online. What I find interesting is that the system and method can be applied to a range of “images”, not just those in a magazine article or a book chapter.
Baby step or not, US7,580,568—filed in 2004—is now a patent. The plumbing and logic for the disclosed system and method have been in operation since late 2002 or early 2003. How the toddler has matured!
Stephen Arnold, August 31, 2009