This Ralph Lauren situation just keeps getting worse... and I can't look away.
First, Ralph Lauren photoshopped a model until she looked like a stick figure.
Then they attempted to
silence criticism by filing
DMCA takedown notices.
And now? It turns out that the model, Filippa Hamilton, was fired by Ralph Lauren a few months ago (before this photoshopped image was created). They terminated her because
she was too fat.
Now, just for clarity, Filippa is 5'10" tall and weighs about 120 lbs (1.77 meters, 54.4 kg). She wears a size 4. That isn't fat. That's malnourished. According to
Health Check Systems,
Weight Watchers, and the
National Institute of Health, she should weigh somewhere between 130 and 170 lbs.
Acceptable Modifications?
In my
last blog entry I identified other forms of digital manipulation performed by Ralph Lauren. Specifically, they digitally alter the clothing on their site in order to sell more products. That is false advertising. That is fraud.
Loyal reader Invincible Spatula made an
excellent observation. In my entry, I pointed out an example image where Ralph Lauren had changed the shirt's color to a very computer generated (fake) blue color. Invincible Spatula had the following comment:
The shirt you show above seems an outrageous example.
What do you think about this tamer example: HTTP://www.levenger.com/PAGETEMPLATES/PRODUCT/ Product.asp?Params=Category=15-765|Level=2-3|pageid=2398
When clicking on the color the picture changes to show the product in the corresponding color. This is the same image manipulation, but doesn't seem as blatantly egregious as your above example.
This seems like an appropriate use of color shifting to represent a product.
Can you confirm that the images are so altered?
Can you comment on your impression with the appropriateness of the alteration? i.e. does it seem untruthful?
Is there a place for image alteration in advertising communication at all? Would you draw a line between appropriate and inappropriate?
I meant to include a 'please' up there somewhere, but I already typed it all. Sorry. Much thanks for your work and perspective.
The Distinction
Invincible Spatula beings up some excellent questions. In this case, the
web page shows a wallet in three possible colors. Users are allowed to select the color and see it applied to the image.
So how is this wallet example any different from Ralph Lauren's shirts? The short answer: As far as I can tell, the default black wallet appears to be real. The red and brown are digital modifications. In the case of Ralph Lauren, nothing is original.
The big differences between this wallet and Ralph Lauren's blue shirt:
- With the wallet, the artist appears to have made a conscious effort to select realistic colors. The color range of each wallet picture appears to have a realistic gamut compared to what I would expect from a digital photograph. (The red and brown still test as digital enhancements -- they grossly fail the color density distribution test. But at least the have realistic color spaces.)
In contrast, all of the Ralph Lauren shirt examples use unrealistic colors. Based on the edging from the white shirt, I think the original color may have been a pale blue -- not a color available for purchase.
- With the wallet, at least one picture appears to be real (the black one). We can check this by subtracting any two images and creating our own mask for the image.

See that black line between the two red items? That is a shadow from one object onto another. The red mask does not include the shadowed area. The color inside the shadow is from the original image. In this case, it is a shadow on the black leather. In all of the images, the shadowed area matches the black wallet's color. (If the brown was original then the shadow should have brown undertones. It doesn't.) So the black wallet is likely original. In addition, the black wallet is the default image and the only image you can zoom-in on.
As an aside: given the mask, I can create my own wallet colors. How about green?

With Ralph Lauren's shirt, every image -- including the default image -- was digitally modified. I can find no image that is "original". We don't know what the original looks like. Ralph Lauren also permits zooming into the digitally enhanced shirt, yet they provide no statement about it not being real.
- With the wallet, the artist only altered the color of the wallet. Everything else -- the spiral notebook, pen, and blue desktop -- appear unaltered.
With the Ralph Lauren shirts, the artist enhanced other portions. For example, in the blue/green shirt example (which features the "fat" Filippa Hamilton who no longer works at Ralph Lauren), Ralph Lauren artists enhanced her belt colors and blurred the background. With the blue shirt example, they touched up her face and eyes, enhanced the belt, made her right arm (photo-left) thinner (look at her bicep), and even altered her armpit. Yet for some reason, they left her shadow on the white wall (see the luminance gradient).


The luminance gradient should have the same bumpy pattern across the entire image. The only exception is the background, which is a uniform color (bumps got smoothed out by the JPEG algorithm.) Areas without bumps or with blurs were clearly manipulated, as are sharp, crisp edges. This includes her face (huge blur on her chin), right arm (photo left), right armpit area up to her shoulder (different pattern compared to the shirt), and her belt near the specular reflection. Even her waist was brought in -- you can see the swirs in the bumpy pattern that show the direction and magnitude of the drag. And these are just the items that I can readily detect. Other forms of manipulation may be undetectable due to scaling and the image's low quality.
Frankly, if Ralph Lauren wanted to use a cartoon, then I wouldn't mind. However, they took steps to make a caricature out of the model and digitally alter their product, and then presented it as if it were real. This is the component that is fraudulently misleading.
What is Acceptable?
I can fully understand the desire to digitally alter colors in order to demonstrate a range of colors without photographing and hosting each independent combination. Imagine if they had 4 configurable options with 10 choices each... that would require 10,000 different images. I can also understand wanting to make this interactive. However, there are ways this can be done without appearing to mislead:
- Include at least one real, unaltered image. This image should be the default image and identified as real. (With the wallet, I believe that the default black wallet is real -- even if it is unlabeled. With Ralph Lauren, none of the shirts appear real; all have been digitally altered.)
- Make the swatches that the user chooses from real, even if the combined example image is an enhancement. (If you have 10 different color fabrics, then include 10 real photos of swatches.)
- Use a color matching algorithm. Many companies do this very well. Sherwin-Williams (paint company) has a color-match system where you can bring in any swatch and they will determine the correct color match. What they don't advertise is that they can also tell you the R,G,B value of the color. (If you're on a budget, then just take a photo and choose a color near the median of the color histogram.)
Now that you have a realistic color for the baseline of the image, you can measure the intensity changes over the original image's base color and alter the new base color accordingly. (This is a chromakey replacement.) The result is much more realistic than having an artist arbitrarily choose a blue that they think looks pretty, even if it does not look like the product.
- Mention somewhere that the combined image is an approximation. With Ralph Lauren's shirt, they simply say "Create your own" and later say "All of our Create Your Own items are personalized and are therefore non-returnable." There is no mention that the item may not look like the image. With the wallet, they come close -- only because the default image appears to be real and the colors are good enough to be realistic.
I remember something you said that I thought was great. It was something like, "the most important tools of image analysis are your eyes and your head." And I think you also said that you don't trust your eyes.
I'm more of the "I see something that does not look like it belongs there" type.
I wanted to tell you that I tried out your error level analysis program. When I first saw it, I thought, "Oh, great. It's going to make a tiny image."
Then, when I opened an image, typed in "95" in the box, and clicked, "WORK!" nothing happened.
Well, not right away.
then, the pixels started appearing.
It took awhile, but it did produce an image the same as the original.
I understand the concept of what it does. What I don't understand are the colors.
Set at "95," the white pixels showed up as black, and the color pixels showed up as blue.
Set at "90," the white pixels turned blue and the color pixels were red and orange.
Set at "50," it's like all of the pixels are orange and reddish.
What do the colors indicate again?
Also, how do you do a luminosity gradient?
Is it a program you wrote?
Can I do it with Photoshop?
I apologize if my questions sound stupid.
Thanks in advance.
PJ
Heh... I didn't even notice that I did that.
Ralph Lauren has user-selectable content. You can select the shirt color and logo pattern. I am directly linking the image from Ralph Lauren's web site. They must have changed the code for the logo pattern.
As an aside, I had downloaded ALL of the different combinations.