Hidden within each photo is a wealth of information about the objects, people, settings, and environment in which the photo was taken. Decades of research in image recognition have enabled the automated labeling and classification of images with reasonable accuracy. While these research efforts have been promising, and some of them have made their way to search engines such as Google and Bing, they have generally not made a dramatic impact on visual search. Not yet. But another decade or two of research will change this.
Using extremely large training data, researchers have been able to create surprisingly simple and accurate algorithms that can determine whether a photo was taken at night or during the day, whether there is a face, or whether it has a sports theme. However, a lower hanging fruit for visual image search, one which could be immediately applied with dramatic impact, involves the manual tags that are already part of most images on social networks such as Facebook and Flickr.