Faced with an increasing backlash against portrait retouching, researchers have crafted a program to quantify just how much the picture has been tweaked.
Retouching images is a fact of life for photographers, especially those doing portrait work. It can be as little as tweaking color and cropping down, but soon it evolves into fixing blemishes, whitening teeth, then resculpting noses and completely changing the way a body looks. There's been an increasing backlash against heavily editing models, especially women in advertising, for presenting young women with unrealistic body types. Some countries in Europe have pushed forward with legislation to ban retouching all-together. The problem with this is how do you differentiate between tweaking image sharpness and removing most of someone's thighs? Or clearing up some stray hair in the face compared to dramatically altering their spine?
This new research out of Dartmouth compares the original and retouched versions of a photo, and using two different methods creates a single number to judge how retouched it is. In the image above, you can see the two methods: the geometric distortions depicted as a vector field, and the photometric distortions. The former measures reshaping of the image, the latter removing of blemishes and sharpening. By combining information from both of these tests, it produces a single score on a scale from 1-5, which says how retouched an image is — a scale which holds up dramatically well when compared to people's perceptions of how altered the image is.
The theory is that ads could be given a rating, much like movies, which give you a quick and easy way of telling the depth of the reshaping. However, it demands cooperation from the industry, because it requires the original image for comparison. Because of this, there will always be ways of getting around the system, unless you can discover a way of guaranteeing that the original image provided is the RAW file directly from the camera. There is some software that attempts to perform a similar trick using only the final image, but it's remarkably difficult to interpret accurately, where the new software seems comparatively straightforward. Hopefully a combination of the two will allow a better grasp of just how heavily retouching is used, and increase awareness of when and where it's abused.