Recognizing the subject of a photograph is a skill we humans take for granted. Whether it is the Golden Gate Bridge, the Eiffel Tower, or a photo of Uncle Harry, we “know it when we see it.”
For computers, however, the task is exceedingly difficult. Variations in color, shading, angle and perspective mean that the same subject can look vastly different in different images. Only recently have academic research projects begun to make advances in efficient algorithms to solve the problem. Now Google has publicly released the first version of their image recognition software Google Goggles, first in the Android platform, and most recently as part of the latest update to their iPhone Mobile App.
How does it work? Using the mobile app the user initiates an image search and then takes a photo with the iPhone camera (3gs or 4 only since close focus is needed). The image is uploaded to the server and after a short analysis, the results identify the image as a landmark, a work of art, a logo, a book cover, a CD cover, or a piece of text.
How accurate is it? Testing with my own photographs displayed on the iPad showed that many major landsmarks were indeed recognized, but not all.