Virginia Tech® home

On the Record: Humans and artificial intelligence help identify photos from the past

Loading player for https://video.vt.edu/media/1_x42apwch...
Category: research Video duration: On the Record: Humans and artificial intelligence help identify photos from the past
Kurt Luther, Virginia Tech assistant professor of computer science, has developed a free software, Photo Sleuth, that uses human crowdsourcing and artificial intelligence to help identify people in Civil War era photographs. 
[00:00:00] >> We've created a project called Civil War photo sleuth. The goal of this project is to try to identify unknown soldiers in photographs from the American Civil War 150 years ago. This is a hard challenge because millions of these photos survive but we have only identified maybe 10 or 20 percent of them. [00:00:18] Most people don't bother to write their names on the back of these photos. So we've created a website called Civil War photo sleuth, it's a free Web site that anybody can use and this website allows people to actually identify these unknown images. The website combines crowdsourcing with automated face recognition. [00:00:36] So face recognition is used to help narrow down a pool of many possible candidates and then humans look at the visual clues in these images to try to identify useful features that can narrow things down even further. So we're taking advantage of the complimentary strengths of human intelligence and artificial intelligence to try to identify these unknown photographs. [00:00:59] Identifying people in photos from historical periods brings some interesting challenges. These photographs are often damaged from age they might have holes or they might have age related wear they also tend to be black and white so they don't have as much detail as we would expect in modern day photographs. [00:01:19] What's interesting about face recognition technology is it works on old photographs just like it does on new ones. Both cases it's looking for facial landmarks on the faces of these unknown people and it's combining the ratios of those different landmarks to create what's called a face print and everybody has a unique face print and by generating this we can compare it with face prints of known photographs to try to narrow down possibilities. [00:01:47] Right now, we're focused on the challenges of historical photographs but these techniques could also be applied to modern day photographs in a certain context. It's important to understand our history. Nothing is as compelling as looking at these images of people who lived long ago and knowing who they are and knowing their stories by using face recognition technology and combining it with crowdsourcing we can rediscover the identities of these people and learn their stories and in doing so learn more about ourselves. [00:02:19] I think the project's also important in a more general sense. It helps us to see how the complimentary strengths of human intelligence and artificial intelligence can be combined to do something together that neither could do by itself.