Professor: Photoshopping Person's Race Common
ROBERT SIEGEL, host:
We used to say that a picture is worth 1,000 words, but that was back in the analog era. A digital image is worth several thousand after you Photoshop it a few times. The latest instance of this is a Microsoft ad which shows three people seated at a conference table, taking in a PowerPoint demonstration with such gleeful smiles on their faces, you'd think they're actually watching "Madagascar 2" or something like that.
On the right is a woman in a business suit who's wielding the clicker, on the left is an Asian man, white shirt, no tie, collar open. And between them is a black man looking like the oldest and most senior, certainly most formally dressed executive in the room.
Across the picture is the text: Empower your people with the IT tools they need. Next comes the Microsoft Poland version of the same ad as it went viral yesterday. Same woman, same Asian man, but instead of the black exec, there's a white man with the very same suit, same shirt, same tie, same wristwatch, even the same black wrist, just a white head for the overwhelmingly white Polish market.
Microsoft says today, we're looking into the details of this situation. We apologize and have replaced the image with the original photograph. Well, joining us is Hany Farid, who's professor of computer science at Dartmouth College. And professor Farid, first, how common, from what you've seen, is this sort of swapping out of heads to make a different racial impression in a picture online?
Professor HANY FARID (Computer Science, Dartmouth College): It's surprisingly common, in fact. In fact, the notion of swapping out heads dates back all the way to the 1800s, when somebody did it to a photograph of Abraham Lincoln.
The University of Wisconsin at Madison got in trouble a while back when they spliced in a black head into a sea of white heads to show more racial diversity. And just earlier this year, Fun Guide of Toronto did the same thing to try to increase the racial diversity of a cover of their magazine. So, it seems to be a fairly common manipulation, in fact.
SIEGEL: Those were both instances of trying to create greater diversity, as opposed to this case, creating a little bit less diversity.
Prof. FARID: Right, right. It seems to be an equal opportunity Photoshop manipulation.
SIEGEL: Now, you're working on software that would be able to detect images that have been manipulated.
Prof. FARID: That's correct. And in fact, in this particular image of the Microsoft ad, there was a glaring error that they made - in addition to the black hand and the white face - and that is the lighting on the face is completely wrong.
So, if you look at the middle person, the white head that was spliced in, it's clear that the lighting is coming from his right. But if you look at the other two people in the room, there's no such dominant light in there. And one of the very powerful tools that we've developed is one that looks at lighting inconsistencies to determine whether something has been manipulated.
SIEGEL: As you said, there is a rich history to manipulating photographs. And there was a wonderful book a few years ago, "The Commissar Vanishes," about the Soviet practice of eliminating people who'd been politically eliminated from all the images in which they'd appeared.
Prof. FARID: That's right.
SIEGEL: But does this strike you as a different order of effectiveness, what can now be done with digital images?
Prof. FARID: I think the difference is twofold today. One is that the tools to manipulate photographs are now available to nearly everybody. And the other thing is the ability to distribute this material widely and quickly. And those two things have combined to have fundamentally changed our relationship with digital imagery.
SIEGEL: Well, Professor Farid, thank you very much for talking with us.
Prof. FARID: Thank you for your time.
SIEGEL: That is Hany Farid, who is professor of computer science at Dartmouth College.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.