Southern Connecticut State University professor Mohammad Tarik Islam has developed Geo Faces, an algorithm and dataset that can determine where people are from based on photographs of them.
In the first stage of the project, Islam took 1.8 million images from Flickr and tested them to see whether the algorithms could match the photo with a location based on the data and visual markers available in the images. The process was about 26-percent accurate, which is about 13 times better than the approximately 2-percent chance a person would have of looking at a photo and determining where it was taken. Islam then focused the technology on facial cues, with the algorithms attempting to determine the ethnicity, gender, and age of the individual in the photo.
Islam took 40 million images from Flickr, Google, Yahoo, and Bing to see whether the algorithm could accurately match the photos with any of the 173 countries in which they were taken. The test showed the algorithm was about 22-percent accurate. Islam says this type of technology could be used to help organizations identify the ethnic and gender makeup of their most common customers, enabling them to market and cater products more effectively.
From New Haven Register
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA