Secure Computing: Sec-C

Two months ago I wrote a blog entry about a photoshop disaster at Victoria's Secret. The original picture featured a model with a digitally removed handbag. While the handbag was the obvious manipulation, it was not the only modification -- they altered the dress, dress color, and even the model! Since I usually do an analysis like this every few weeks, I thought that after two months nobody would remember it.

With one Twitter post by Tim Oreilly on 27-Dec-2009, my little blog entry went from obscurity to overexposure. The last week has certainly been interesting. (I'll blog later this week about the flash-mob distributed denial of service attack that lasted a week.)

A large number of people wrote to me via email and blog comments, and many people discussed the same issues. In this blog entry, I will be addressing some of the more common queries.

Natural Skin

In my blog entry, I pointed out that the model had been recolored. Specifically, her dark skin had been significantly lightened. A significant number of people replied with comments concerning this statement. The comments ranged from stage lighting and color reproduction issues to questions about whether the recoloring was significant. For example, a comment by Clumpy summarizes the common feedback:

Gotta disagree on the skin color thing - my darker-skinned friends often photograph very differently depending on the lighting. The two photos are well within the realm of possibility, especially when you consider the relatively darker stage of something like a fashion show.

The only good thing about the recoloring is that it was poorly done. Either the artist lacks talent, or had no time to do a quality job. Someone with the time and effort would have recolored all of the skin, and not left patches that contain the original color.




In this case, the artist forgot to recolor two patches. One is at the top of her ear. If this were due to hair in front of the ear, then we would not see the same detail around the curve of the ear.

The second patch is on her neck. The model is in a complex lighting environment. Most likely, they had lights and reflectors around her to illuminate her entire face. This is why her nose, arms, and fabric folds cast no shadow. The second light source is the sun, coming from picture's right and behind her. None of these light sources account for the dark shadow on her neck. The shadow is likely part of her original skin color.

Victoria's Secret did not just lighten her skin, they significantly altered the color.

Why Analyze It?

While most people got it, some had trouble seeing the forest for the trees. A few people commented that all ads photoshop their pictures, that the modifications are not significant, or that they could not understand the point of this exercise. As tgunter commented on Reddit:

I have to say, I was a bit underwhelmed. The only reason this photo is of note at all is the fact that they did such an obviously bad job of removing the handbag.

There are two reasons that I went into this kind of depth.

Reason 1: Everyone's doing it
I've never been a big fan of peer pressure. Just because everyone is doing it does not make it right. We have truth in advertising laws for a reason. If I show you a picture of an item for sale, then you have the expectation that the item will look like the picture. In the US, pictures that advertise food must use a real picture of the food. In the UK, an Olay ad featuring Twiggy has been ruled as misleading. The rationale: if you are selling makeup to make people look better, then don't photoshop the model to make her look better.

A few years ago I had a manager who called this the "Alpo" mentality: eat your own dog food. If your company creates a product, then the company should use the product and the ads should feature the product.

With Victoria's Secret (and Ralph Lauren and Macy's and most other companies), they offer clothing, but show you pictures of digitally modified clothing. The same goes for most other advertisers (if you're selling a radio, then show a real person holding the radio!). At best, this is misleading the consumer. It is time to raise awareness and change the norm.

Reason 2: Technique
It is not so much the image as it is the techniques. There are common clichés on the Internet like the sarcastic "I can tell from the pixels" and the disbelieving "Pics or it didn't happen". However, there is a hidden truth: Given the right tools, you can tell from the pixels, and pictures can lie.

A couple of people mentioned that digital pictures are not permitted in court. How do I say this politely: they're wrong. As digital evidence, it is essential to be able to distinguish between real and digitally altered data.

Of all of the comments, I think a Reddit post by Jericho really nailed it: "The point of the article was not to uncover scandalous alteration being done at Victoria's Secret in a racist or sexist fashion, but to look at the tools and methods one can use to detect alterations."

Tools

By and far, the most common question was about my tools. I've developed my own analysis software, but my tools are not publicly available. Moreover, I have no intention to release my code as open source.

Instead of releasing tools, I describe methods and techniques in my public presentations and blog. This forces people to develop their own systems and that is the best way to learn. If I give away tools without training, then people will use them wrong. But if I force people to develop their own ("teach a man to fish"), then they will learn. Already, three people have implemented variations of the error level analyzer:

• Error level analyser by Noah.
  
• Error Level Compare by Soco Software.
  
• ELA by Schlake.

There are also people working on variations of the copy-move and luminance gradient algorithms. And that is just the beginning -- there are hundreds of algorithms available for performing image analysis.

A Warm Thank You

The vast majority of feedback over the last week was very positive and appreciated. Thank you for the support, and I'll do my best to keep this blog interesting as we start a new year.