Estimating another person's age comes naturally to most people. Some find it more challenging than others, but it's an ability we usually don't give much thought to. For computers, however, it's not all that easy. There have been many attempts at age estimation software, but the one offered recently by a group of researchers from Singapore seems unique. It relies on widely available and inherently diverse web resources, and can therefore be used across all ethnicities.
Bingbing Ni of Singapore's Advanced Digital Sciences Center, along with Zheng Song and Shuicheng Yan from the National University of Singapore, used age-related search queries to automatically mine 400,000 images from popular image search engines such as Google image and Flickr. This technique was also applied to YouTube, with 10,000 video clips selected. The faces in the collected images were then age-tagged and used for training the system. The videos did not undergo the tagging, but were used instead for teaching the system how faces change when viewed from different angles and in different lighting.
UPGRADE TO NEW ATLAS PLUS
More than 1,200 New Atlas Plus subscribers directly support our journalism, and get access to our premium ad-free site and email newsletter. Join them for just US$19 a year.UPGRADE
After removing the images that did not meet quality standards or failed the cross-check of results produced by different face recognition techniques (false alarms), the database still contained 77,000 images with a total of 219,000 faces. This is reportedly the largest set of visual reference materials ever used for age estimation purposes. Thanks to the implicit diversity of this database, the system is able to provide age estimations for members of all races and ethnic groups.
The estimates produced by the system on the basis of regular-quality images are accurate to plus or minus five years. The researchers hope to improve on this result, as well as reduce the amount of errors for lower quality input. To highlight one possible application of the software, they prepared a Vision-based Demographic Advertisement System. As could be expected, this system serves up advertising relevant to a particular age group (think about this the next time you see an advert for denture cleaning tablets).
Fortunately, the list of possible applications doesn't end there. We can see such a system being used for personalizing human-computer interfaces on the go (font size, etc.), age-tagging your photo collection, preventing minors from accessing adult internet sites or even for detecting child pornography.