Sara Ziff: So I read that you’re proposing a software tool that measures how much fashion and beauty images have been altered on a scale of 1-5. How did you start working on this?

Eric Kee: In general, I’m interested in problems of whether a picture is real or not. And that’s what my advisor Professor Hany Farid does. I began working with Hany about four years ago now and we’ve worked on a variety of projects.

SZ: What brought you to this particular project — measuring the degree to which fashion and beauty images have been altered?

EK: What brought us to this was the social problem that photographs of people are modified to make them look more “attractive,” and there are documented health issues that relate to body image. We were aware that people were trying to do something about this and we thought we could potentially contribute. Hany heard that in the UK they wanted to legislate that modified photos should be labeled and immediately we realized that almost every photo is going to be modified in some way. It’s not going to be useful to just put a label saying whether or not the photo was modified because every photo would have a label on it, which is meaningless. And so that’s when we came up with this idea to devise a rating system to determine just how much or what kind of changes were made to photographs.

SZ: There is, in fact, research that shows that highly idealized, digitally altered images promote eating disorders, right?

EK: There is medical research — psychological research — that shows this […] We cite a number of these articles in our paper […] So that research has been out there. I think what’s been missing is the idea that maybe you can actually measure the amount of change that has happened [in the retouching process].

These photos demonstrate the way software designed by Hany Farid and Eric Kee rates images according to how much they have been digitally altered in post-production. The program uses an algorithm to assign images a number on a scale of 1-5, with 1 being the lowest level of alteration and 5 being the highest. Above: from left to right, these images rate 1.4, 2.4, 3.0, 3.5, and 4.3. Below: the unretouched shots, for comparison.

SZ: What’s an example of a photo that rates as a one versus, say, a five?

EK: Well, almost every photograph is retouched to some degree. Benign changes are changes that would alter the brightness of the photo, or maybe adding a little more contrast, or changing the color of the photo overall — those types of things. Or perhaps if you have a photograph on film and you want to remove [grain] — things like that. Those are superficial changes, so you will get the results of the retouching, but basically the person in the photo doesn’t appear changed, just the quality of the photograph itself. On the other hand, when you look at more extreme forms of retouching, they can do things like remove wrinkles from peoples’ faces, which is one of the most popular, and make subjects look slimmer or more muscular.

SZ: My friend, the model Crystal Renn suffered from an eating disorder as a straight-size model. She did a photo shoot in 2010 where she was digitally altered to the point where she looked like her former emaciated self, so she took the photographer to task.

EK: I’ve heard about this.

SZ: So, well, one of the things that fascinates me about your work is the suggestion that models, with the help of your software, could maybe one day say, “I don’t want to be a 5 on the Photoshop scale, I want to be a 1.” And I thought that was cool because I see this as an important step for models as workers — as so-called independent contractors — to have some control or final say, or at least some significant input into the final product of their work.

EK: Hany and I think that would be really cool. There are a lot of social things that have to happen in order to actually bring that into being. But fundamentally, technologically, it’s possible. Prior to doing this project, it was not clear that it would be possible to measure a person’s opinion of how much someone’s appearance [in a photo] had changed. Think about if you look at a before and after: it’s difficult to put into words how much a person’s appearance has changed. It’s difficult for a person to do, and it was unclear whether that was something we could actually measure and quantify with a computer algorithm. But we showed that you can actually quantify that. And the advantage of measuring something like this is that now we can talk very concretely about how much has a photograph changed. And I think it’s really exciting that potentially models and other groups and individuals could specify just how much [their photo should be altered], and the language would be this kind of rating system, which formalizes this feeling about how much the photograph has changed.

SZ: Have you heard any response from the fashion or beauty industries? Has anyone reached out to you and said, we want to adopt this?

EK: We have received some inquiries from people in reporting roles where their audience was the beauty industry, but I have not received emails personally from a photo retoucher or modeling agencies or things of that nature, and certainly not companies that sell beauty products. Evidently they do some of the heaviest retouching, which is bad for the other reason that not only are there body image issues, but the there’s also the issue of truth-in-advertising in the cosmetics industry.

SZ: Legislators in Britain, Norway, France and the U.S. have said they want digitally altered photos to be labeled. And I know that Seth Matlin, the founder of offourchests.com, is the driving force behind the “Media and Public Health Act,” which aims to make consumers in the US more aware of the digital enhancement that occurs in advertizing by instituting truth in advertising labels. In your view, what’s the likelihood of this happening? *

EK: I’m probably not qualified to comment on that. There are a lot of people who are very interested in instituting these labels, but I don’t know what the likelihood of them adopting something like what we’ve designed is. When we published this work, I thought this was important. The media response that we’ve received has really been overwhelming.

“Retouching a photo is a very slow process. You do it gradually — one change at a time — and it’s not hard to stray from reality slowly.”

SZ: How would this software be implemented?

EK: I think it depends on a lot of the social factors. One idea is that some kind of rating or standardization organization — like an organization that assigns a movie rating of PG13 or R — could use our software to assess photos. On the flip side, it might be possible that the photo retouchers themselves could use the software to see, well, if I do this or I do that, it’s going to increase the rating. That could be very beneficial to retouchers when they don’t actually intend to do something nefarious. I think that it’s very challenging when you’re retouching a photo because it’s a very slow process. You do it gradually — one change at a time — and it’s not hard to stray from reality slowly. You might make a degree of alteration that you didn’t intend. So the second application would be to have the ratings available to all retouchers.

SZ: You said that this software could be implemented by a rating organization, like the Motion Picture Association of America rates movies. Is there any such organization in existence that could rate fashion and beauty images?

EK: Not to my knowledge. I’m getting educated on all of this very quickly now that this is in the public domain. There are a lot of stories in the media — every day now — about people wanting for this to be legislated, really pushing hard, especially in the U.K. and France. And if there’s enough social momentum, it seems plausible that some kind of social ratings organization could be created. As a model, I assume it would be exciting for you to be able to mandate, oh you can’t do this or that to my photos.

* Since this interview, Israel became the first country to pass a law that requires advertisers to label photos that were manipulated to make a model appear thinner.