The Shape of Days

Heavy sigh. This Adnan Hajj story just keeps getting worse and worse. At a quarter of nine tonight, Rusty Shackleford of The Jawa Report e-mailed me to say that he’d found another retouched photo of Hajj’s in the Reuters archive. The picture was certainly mis-captioned; it purports to depict an Israeli F-16 firing “missiles,” when that’s definitely not the case. Here’s the photo in question:

And here’s a link to the photo, with caption, on the Reuters site.

Now, you don’t have to be a rocket scientist to recognize that those aren’t missiles coming off that F-16. They’re flares, countermeasures used against heat-seeking anti-aircraft defenses. Photographs of F-16s dropping flares can be found easily on the Internet.

The Hajj photo also seems to show three small objects below the F-16 and ahead of the three bright flares. What these are, I can’t say, and won’t speculate here.

My focus is on the flares themselves, particularly that third one from the left. Rusty e-mailed to say that he thought that flare had been added with Photoshop or a similar tool, that it was just a copy of one of the other two flares pasted in. At first I didn’t agree. I told Rusty, via e-mail, “I’m going to stay out of this one because there just aren’t enough pixels there to be absolutely certain. Between the low resolution and the artifacts introduced by JPEG compression, it’s impossible to be 100% sure that the photo has been retouched.”

Rusty replied that he thought the third flare was a pixel-for-pixel copy of one of the other two. He was pretty insistent. So I gave it another look.

You know what? I think he’s right.

Yesterday I did some quick-and-dirty analysis on another Hajj photo using something called a difference matte. See, each pixel in a digital photograph is represented by numbers representing color values. But there’s nothing special about the numbers themselves. Like any numbers, they can be subjected to the torture of simple arithmetic. For example, you can subtract the color values of one pixel from another pixel. If the two pixels are identical, they’ll have the same color values, so when you subtract them you’ll get zero. Since zero is the computer’s way of representing black, the difference of two identical pixels is a black pixel.

If you do that operation on every pixel in a picture, the resulting picture is called the difference matte. If the two pictures are very different, the difference matte will be a riot of colors. If they’re very similar, the difference matte will be very dark. If they’re identical, the difference matte will be solid black.

In other words, pulling a difference matte gives you an objective, obvious way of determining how similar two photos — or parts of photos — are.

So I did a little analysis on the photo Rusty sent me.

The first step was simply to crop and enlarge the portion of the picture I was interested in. This is trickier than it sounds. There are several different ways to enlarge a photograph digitally. Most of them involve interpolation, which is a technique for figuring out what color the new pixels should be. Interpolation techniques have gotten pretty good over the years; you can now enlarge photos a bit without making them look like they’ve been enlarged.

But that’s not what I wanted to do. I didn’t want to alter the image data in any way; I just wanted to see it bigger. So I turned off interpolation and created this image:

Then I drew a rectangular selection around the first of the three flares — for sake of brevity, I’m gonna refer to them as flares 1, 2 and 3, numbered from left to right. I selected flare 1, as seen here:

Then I superimposed a copy of that selection over flare 3, as seen here:

Then I told Photoshop to show me the difference matte — literally, the arithmetical difference between the two regions of pixels. This was the result.

As you can see, there difference matte is mostly black. This shouldn’t be surprising, though, because the vast majority of the pixels in my selection were blue sky, and the colors of those pixels were all very close to each other. So they canceled out and became black.

But if you look closely, you can still see the outline of the flare and its smoke trail. It’s dim, but it’s there. Which means that while the colors of the pixels in the two samples were similar, they weren’t identical.

Then I moved on to flare 2.

Ah, very different. Whereas the outline of the flare and its smoke trail were dim but discernible in the first matte, in the second matte they practically vanish. There’s a ring of very dim pixels (with RGB values of around 20) around the flare itself, but the smoke trail is all but invisible.

There is some noise in the upper left, however. That’s because my copy of flare 2 slightly overlapped the smoke trail from the actual flare 2, which meant I was subtracting white smoke from blue sky. So we see something there.

(Rather than going back and re-drawing my selection more carefully, I decided to leave it in. Mostly because I’m lazy and I didn’t want to do all the selecting-and-copying-and-pasting work over.)

Now, none of this means much. We’re looking at two mostly-black rectangles, one slightly blacker than the other. So to more clearly see just how these two mattes compared, I applied an enhancement to them. I mapped the RGB value of white (that is, 255) to an RGB value of 40. That has the net effect of making everything that’s almost black brighter, so we can more clearly see things that are very dim.

Here are the results.

As you can see, the first difference matte — between flare 1 and flare 3 — when enhanced lights up like a Christmas tree. Differences in color values that were very small are magnified so we can see them clearly.

But the second matte — ignoring the noise in the upper left — is still very, very dim. Even under enhancement, the outline of the smoke trail is difficult to see, and in some places almost indistinguishable from the background noise.

Now, I want to be clear about this. None of this is scientific, and none of it should be considered conclusive. This is a really low-resolution photo we’re dealing with, and it’s been put through a JPEG compression algorithm that would make a strong man weep. It is simply not possible to say with 100-percent certainty that the third flare is a copy of the second flare.

But I’m convinced.

I think this is the second example yet found of a photo taken by Adnan Hajj that was drastically retouched to change its content before it ran on the Reuters news wire.

Thanks, of course, to Rusty for calling my attention to this in the first place, and for keeping my attention on it when I wanted to weasel out.

• Adnan Hajj suspended over doctored Beirut photo
• A photojournalist weighs in on the Adnan Hajj scandal
• The Adnan Hajj story is getting bigger