Without a Crutch
Thursday, 29 September 2011
Over the last 48 hours I have received nearly a dozen requests to voice an opinion on an alleged fake photo. The photo, by Reuters photographer Anis Mili, is described as "A rebel on crutches fires a rocket propelled grenade while fighting on the front line in Sirte September 24, 2011". The photo, taken from right behind the RPG as it is launched, is truly amazing.
There's the old saying that we, as skeptics and people constantly bombarded by digital manipulation, take to heart: If something it too good to be true, then it probably is. This picture is just too good. It immediately led to a thread on Reddit and on various blogs, where people questioned the authenticity and claimed that it is fake. There were a variety of arguments posed by other people:
Since enough people have asked, here's my core findings. (For the record, I ran many more tests than the ones listed here, but I'd rather not write a 20-page report if I don't need to).
Second, can a guy on crutches -- with a metal leg brace -- fire an RPG? Having never fired one myself, I asked a friend of mine who has fired one. The short answer is "yes". Apparently, an RPG's firing mechanism is used to aim, launch, and steady. Since the back end is open, there is no kickback. (Guns have a kickback because the equal-and-opposite reaction to the bullet going forward is redirected sideways. Since RPG launchers don't redirect the energy from the rocket, there is no kickback.)
As far as catching the backwash goes, it is definitely possible. Over the course of a half second, the flame starts from nothing and grows to a huge fire. Thus, it could be captured at just the start. As far as catching it manually goes, the odds are slim. However, we don't know what type of camera was used. If he was taking photos in rapid succession, or using a flash sensor to catch the exact moment, then it is very possible.
And the shadows? If you look down the road, you will see that he is standing on a slope. Since the road is a nearly uniform color, we have no real sense of depth. We don't know how the road is sloped where he is standing. Shadows can do really cool things on slopes -- they can appear to vanish, be too long or not long enough, and even appear to bend in the wrong direction. It all depends on the slope. There is a huge conspiracy about the Apollo moon landing because some people think that the shadows on the uneven lunar surface are wrong. Similarly, we know that the road is uneven so we cannot trust the shadows. Having said that, there is a faint shadow directed from both crutches (look hard for the left one since it is only partially visible). The front of the RPG launcher is likely on the person and the back appears as a thin "<" in his shadow.
Finally, was he really shooting at nothing? While there may be nothing on the street, look just past the sidewalk, at about the 2:00 position from the RPG. That's not a building or a fountain; that's a friggin' tank! It's faded from haze and blurry from the depth-of-field, but it sure looks flat with a turret on top. Over his left shoulder are other pieces of equipment. We don't know exactly what he is aiming at because of the haze, blur, and the fact that his head is in the way, but it certainly does look like there is something being targeted.
The facts represented by the picture check out, so now we can focus on digital photo analysis.
For clarity: just because the picture was saved using Photoshop does not mean it was intentionally manipulated or modified. If someone wants to save a picture at a size that works on the web, they are just as likely to choose Photoshop as any other program.
Photoshop has some undesirable side-effects, such as minor color enhancements, automatic sharpening, and constant JPEG degradation. Simply using Photoshop to save a picture will result in unintentional modifications.
Error Level Analysis (ELA) identifies regions of the picture that are at a different error level potential. In general, everything should be about the same color. However, simply saving with Photoshop is enough to increase the error level potential along high contrast edges (like the crutch, part of his clothing, the bottom ground, and even the flame from the rocket).
In the case of the ground, there are black specs in the white, washed-out region. That makes the entire area a high contrast region. Combined with Photoshop's focus on emphasizing high contrast regions, the area gains a higher error level potential. In contrast, the RPG, flame, and shockwave don't stand out at all under ELA. So there is nothing suspicious here.
Luminance Gradient (LG) focuses on light direction and noise from the camera's sensor. The entire picture should have a bumpiness pattern to it. In this case, we have the bumpy pattern, and it even exists in the flame and shockwave. If this were computer graphics or a splice, then it shouldn't have the same bumpy pattern. The only places without the bumpy pattern are the crutch and the shadow -- and we've already identified that those were automatically sharpened by Photoshop.
The high-frequency noise test identifies the peaks and valleys in the two-dimensional high-frequency signal. With a real picture, it should look like a random set of dots. You won't be able to identify anything. (E.g., you can't look at this dot pattern and identify the trees, road, or even most of the person.) Sharpening, like what Photoshop does automatically to JPEG images, will result in a dense cluster of dots -- like we see on the crutch and over the flame. If you cover the crutch with your finger, you cannot tell where the background ends and the person begins. Similarly, the RPG and shockwave are completely unidentifiable. This looks real.
If we go through this entire exercise, then there is really only one intentional modification that I can identify. It is in the RPG's flame -- someone made it a little brighter. A little, but not a lot. And I don't think this violates Reuters guidelines for enhancements. (But the final judgement is up to Reuters, and they already called the picture one of the Editor's Choice photos.)
Typically, amazing photos come about through digital modifications. However in this case, Anis Mili has truly captured an amazing photo. And he did it without using a crutch like Photoshop.
And now for my only criticism... As I understand it, there are two places where you don't want to stand when someone is firing an RPG. First, you don't want to be in front of it (duh). And second, you don't want to be directly behind it because that back-blast can kill. Anis Mili: What were you thinking?!?
I'm sure someone is going to ask for my tools. My tools are not publicly available. The tools are nothing more than a microscope; anyone can look through a microscope, but you won't know how to interpret the results without training. And as the DC3 wrote in their evaluation of the tool, there is a very steep learning curve.
Special thanks to Chris Hanson, Ms. A., Coredump, Drew Hallowell (my first twitter follower!), and everyone else for forwarding this to me.
There's the old saying that we, as skeptics and people constantly bombarded by digital manipulation, take to heart: If something it too good to be true, then it probably is. This picture is just too good. It immediately led to a thread on Reddit and on various blogs, where people questioned the authenticity and claimed that it is fake. There were a variety of arguments posed by other people:
- The RPG would knock over a guy on crutches.
- The shadows are missing.
- He's not firing at anything. (What's he trying to do, kill trees?)
- The back-blast from an RPG is bigger than that.
- The RPG flame has a blurry edge and was pasted.
Since enough people have asked, here's my core findings. (For the record, I ran many more tests than the ones listed here, but I'd rather not write a 20-page report if I don't need to).
Step one: Observation and Identify
Starting at the beginning, the picture claims to show a fighter in Libya, in Sirte, on the front lines, on September 24, 2011. So the basic question: was the front line there on that day? According to news reports, that is exactly where the fighting was located on that day.Second, can a guy on crutches -- with a metal leg brace -- fire an RPG? Having never fired one myself, I asked a friend of mine who has fired one. The short answer is "yes". Apparently, an RPG's firing mechanism is used to aim, launch, and steady. Since the back end is open, there is no kickback. (Guns have a kickback because the equal-and-opposite reaction to the bullet going forward is redirected sideways. Since RPG launchers don't redirect the energy from the rocket, there is no kickback.)
As far as catching the backwash goes, it is definitely possible. Over the course of a half second, the flame starts from nothing and grows to a huge fire. Thus, it could be captured at just the start. As far as catching it manually goes, the odds are slim. However, we don't know what type of camera was used. If he was taking photos in rapid succession, or using a flash sensor to catch the exact moment, then it is very possible.
And the shadows? If you look down the road, you will see that he is standing on a slope. Since the road is a nearly uniform color, we have no real sense of depth. We don't know how the road is sloped where he is standing. Shadows can do really cool things on slopes -- they can appear to vanish, be too long or not long enough, and even appear to bend in the wrong direction. It all depends on the slope. There is a huge conspiracy about the Apollo moon landing because some people think that the shadows on the uneven lunar surface are wrong. Similarly, we know that the road is uneven so we cannot trust the shadows. Having said that, there is a faint shadow directed from both crutches (look hard for the left one since it is only partially visible). The front of the RPG launcher is likely on the person and the back appears as a thin "<" in his shadow.
Finally, was he really shooting at nothing? While there may be nothing on the street, look just past the sidewalk, at about the 2:00 position from the RPG. That's not a building or a fountain; that's a friggin' tank! It's faded from haze and blurry from the depth-of-field, but it sure looks flat with a turret on top. Over his left shoulder are other pieces of equipment. We don't know exactly what he is aiming at because of the haze, blur, and the fact that his head is in the way, but it certainly does look like there is something being targeted.
The facts represented by the picture check out, so now we can focus on digital photo analysis.
Step Two: Identify the picture's baseline
According to the metadata and the JPEG's ballistic profile, it was created using Photoshop "Save As" with a quality level 8. (When you save with Photoshop, you are asked to select the quality -- from 0 to 12. In this case, the person chose "8".) A photoshop image at "level 8" is roughly equivalent to a standard JPEG at about 94%.For clarity: just because the picture was saved using Photoshop does not mean it was intentionally manipulated or modified. If someone wants to save a picture at a size that works on the web, they are just as likely to choose Photoshop as any other program.
Photoshop has some undesirable side-effects, such as minor color enhancements, automatic sharpening, and constant JPEG degradation. Simply using Photoshop to save a picture will result in unintentional modifications.
Step Three: Profit (err... I mean "Analysis")
So let's run the picture through some analysis methods...Error Level Analysis (ELA) identifies regions of the picture that are at a different error level potential. In general, everything should be about the same color. However, simply saving with Photoshop is enough to increase the error level potential along high contrast edges (like the crutch, part of his clothing, the bottom ground, and even the flame from the rocket).
In the case of the ground, there are black specs in the white, washed-out region. That makes the entire area a high contrast region. Combined with Photoshop's focus on emphasizing high contrast regions, the area gains a higher error level potential. In contrast, the RPG, flame, and shockwave don't stand out at all under ELA. So there is nothing suspicious here.
Luminance Gradient (LG) focuses on light direction and noise from the camera's sensor. The entire picture should have a bumpiness pattern to it. In this case, we have the bumpy pattern, and it even exists in the flame and shockwave. If this were computer graphics or a splice, then it shouldn't have the same bumpy pattern. The only places without the bumpy pattern are the crutch and the shadow -- and we've already identified that those were automatically sharpened by Photoshop.
The high-frequency noise test identifies the peaks and valleys in the two-dimensional high-frequency signal. With a real picture, it should look like a random set of dots. You won't be able to identify anything. (E.g., you can't look at this dot pattern and identify the trees, road, or even most of the person.) Sharpening, like what Photoshop does automatically to JPEG images, will result in a dense cluster of dots -- like we see on the crutch and over the flame. If you cover the crutch with your finger, you cannot tell where the background ends and the person begins. Similarly, the RPG and shockwave are completely unidentifiable. This looks real.
If we go through this entire exercise, then there is really only one intentional modification that I can identify. It is in the RPG's flame -- someone made it a little brighter. A little, but not a lot. And I don't think this violates Reuters guidelines for enhancements. (But the final judgement is up to Reuters, and they already called the picture one of the Editor's Choice photos.)
Photo Finish
By using a suite of analysis methods, it becomes extremely difficult for a fake image to pass unnoticed. While an intentional forgery might pass one or two tests, it takes a level of skill that most photographers and amateur graphic artists lack. This picture easily passes every test (including a whole slew that I didn't include here). I have no reason to question the authenticity of this picture.Typically, amazing photos come about through digital modifications. However in this case, Anis Mili has truly captured an amazing photo. And he did it without using a crutch like Photoshop.
And now for my only criticism... As I understand it, there are two places where you don't want to stand when someone is firing an RPG. First, you don't want to be in front of it (duh). And second, you don't want to be directly behind it because that back-blast can kill. Anis Mili: What were you thinking?!?
I'm sure someone is going to ask for my tools. My tools are not publicly available. The tools are nothing more than a microscope; anyone can look through a microscope, but you won't know how to interpret the results without training. And as the DC3 wrote in their evaluation of the tool, there is a very steep learning curve.
Special thanks to Chris Hanson, Ms. A., Coredump, Drew Hallowell (my first twitter follower!), and everyone else for forwarding this to me.
And, regarding the "RPG casts no shadow on the ground" counter argument - actually the shadow is there. Use your favorite software to stretch the guy's shadow far enough to make up for the flat visual angle, and the obviously elongated object at about the area where his shoulder's shadow is pops right out at you, exactly where it's supposed to be, at plausible proportions, in line with the sun angle relative to his body, and whatnot.
Take a look at one I did a few years ago:
http://www.hackerfactor.com/blog/index.php?/archives/322-Body-By-Victoria.html
You say in the high frequency noise test if one were to cover the crutch one could not tell the background end from the subject.
I'm looking at this on my iPhone and can ABSOLUTELY see the outline of the subject when I cover the crutch. I can see the curve of his head, shoulders and back and even clothing details. Does anyone else see this?
Not tying to debunk the analysis. I'm just not seeing what your text says I should be (or more accurately I am seeing what it says I shouldn't.)
.
What you are seeing are the high-contrast areas that photoshop automatically sharpened. Notice how those edges that you mention also appear in the ELA as a higher error level potential.
If you exclude the areas automatically modified by Photoshop, then you exclude a few of the edges around the person. Excluding those, there is no distinction between the head, blast, shockwave, clothing, street, trees, and the rest of the non-auto-photoshop regions.
Someone doesn't know the first thing about rockets.
* The shadows are missing.
Someone hasn't heard of perspective and never seen a shadow on a rough surface at a steep angle.
* He's not firing at anything. (What's he trying to do, kill trees?)
Because if we fake a photo professionally like that, we don't bother with plausibility but instead pick the most generic backdrop imaginable.
* The back-blast from an RPG is bigger than that.
This has to be the most stupid one of the lot. Clearly the blowout is in the early stages shortly after ignition.
* The RPG flame has a blurry edge and was pasted.
See previous entry.
This list reads like a bunch of moon-landing denier arguments. Why do you even bother.
You seem to be assuming that this launch is completed and that either the launch is a slow process, or that the entire image was captured instantaneously. All are incorrect assumptions.
First, this picture shows the very first microseconds of the launch process. You don't see the RPG in the air because it hasn't left the launcher yet. The shadow from the front of the launcher is likely on the person, so you won't see it as an independent shadow.
Second, we can't see the type of RPG and launcher. Depending on the type, the launch will take anywhere from a tenth of a second to a half second. (Think about the time it takes to launch one of those boyscout model rockets. An RPG is much faster.) So we're talking about a few microseconds to capture this photo.
Finally, digital cameras -- even fast ones -- have a scan rate issue with extremely fast objects. Moreover, the camera may scan from bottom to top, right to left, or even a split pattern. The scan order depends on the chipset and electronics. A good example of this appears when you take a digital photo of a propeller or fan: http://www.peteconnolly.co.uk/blog/wordpress/2007/02/20/strange-propeller-effect/ (Things that go in circles enter and exit the scan row faster than the row can be scanned.) You can see the same effect if you put your hand on a photocopier and move your hand while the page is being scanned.
In this case, the camera appears to be scanning from bottom to top. I say this because the top of the explosion is a little larger than the bottom. Moreover, if you measure the distance from the middle of the explosion (assumed to be just above his shoulder) to the top and bottom, then you can estimate the expansion rate. With my rough measurements, I get about 44 pixels from center to bottom and 50 pixels from center to top. That means the explosion is expanding at a rate of about 6 vertical pixels per 94 scan lines. (Keep in mind, these are rough estimates.)
Now, his feet are about 420 pixels below the center of the explosion. So when the camera scanned his feet, the explosion was about 67 pixels tall (94 - 420 x 6/94). His exposed left ankle (the healthy ankle, not the broken one in the brace) is about 26 pixels tall. So we're talking about something that is less than 6" tall. [I think I got the math right, but feel free to correct me if I got it wrong.]
Now we need to consider the observed angle of the shadows. The back end of the RPG launcher is about 2 feet behind him and it casts a shadow that is about 6 pixels long. So an explosion that is not more than a few inches in diameter is not going to cast any discernible shadow. Thus, we don't see a shadow from the explosion.
As an aside, the "expanding while being photographed" also explains why there is more stretch/blurry above the explosion and not under it. We've basically viewing a Doppler shift; the bottom is compressed and the top is expanded.
So in regards to your attempt to debunk this real photo: thanks for playing. I'll give you a B- for effort.
What would be the effect of superimposing the 'flash' of the alleged RPG taken with the same camera (ie accounting for noise, etc.) but at a different location/occasion, and possibly even that position (x,y) on the exposure?
Great question.
Let's go all out and assume same exact camera, same resolution settings, same focal distance, and same exposure settings (since those all impact the noise levels). The noise leve would look the same.
Except... for two problems.
Problem #1: How do you combine the two pictures?
If you just do a splice (no blending), then you'd have a hard-edge where the colors don't match. (That would fail a huge number of tests.) Since we don't see a hard edge, that isn't the case here.
The other option is to blend the pictures. However, blending changes the noise levels along the blend. So that didn't happen either.
Problem #2: How do you retain the overall error level potential of the image?
If one picture is spliced onto another, then the spliced region has a higher error level potential -- either the whole region or the edge around the region. This will show up under ELA (unless you splice along the JPEG grid), PCA (regardless of whether you splice along the JPEG grid), LG, as well as other tests. In this case, the HF noise might look right, but LG, PCA, and other tests will still detect the modification.
(As an aside: I see this type of thing often. It is common for me to receive a picture where someone says "His head looks odd" and it turns out that the head is fine and its the arms that are wrong.)
In this picture, there is no indication of splicing or blending.
It looks as though his left arm is down and perhaps holding onto the other crutch instead of up and holding on to the RPG (as noted by the elbow inflection).
My thoughts were that perhaps he was using the crutch as a stabilizer for the RPG but you can see by the angle of the left crutch that he's not using it for the RPG.
Just my .02
Really? Are you sure? Because I don't see a left arm. I see a sleeve, but that type of clothing has loose sleeves. His left arm is not visible -- it very well could be supporting the RPG.
Then again, it looks like -- on some RPGs -- the left hand doesn't hold any weight and it just for a little steadying. Example video:
http://www.military.com/video/rockets/rocket-launchers/rpg-test-fire-in-afghanistan/665851878001/
And with some RPGs, the left hand doesn't even need to be all the way over to the right:
http://cheaperthandirt.com/blog/wp-content/uploads/2011/07/ASpurlin_RPG.jpg
However if you look at the left sleeve, even with loose clothing, the angle at which it drops indicates that his arm is down and not over to the right (in my opinion).
Other questions I have when looking at it, why does his right sleeve appear to be short (because you do see his elbow), but not his left? It does not appear to be rolled up.
If his left hand is indeed steadying the RPG, then why is there a crutch in front of him at that angle. He's just resting it on his chest? Makes little sense to me.
I like your forensics work though.
Your photo suggests that the furthermost combatants on a day when all hell was breaking loose was a casually posing half cripple with no cover, obviously mad, and an equally mad photographer.
The photo without pictorial analysis damns itself
Dude, this is a country where the leadership has turned its military against its own people, where entire towns are slaughtered for not supporting the government, and where the "leader" (Kadafi) seems to be a suffering from long-term dementia. And you're talking about "common sense"?
Think of it this way: if you were fighting for your freedom and for your right to live, would you let something like a leg wound and little combat training stop you from rebelling against the government?
And as far as being a war reporter and photographer on the front line -- they are either very brave or very crazy (or both). But I'm glad that someone is documenting this.
Also, I just don't buy the slope changing the shadow thing. In this instance, there isn't that much slope, and there is not the slightest indication of a shadow from the RPG on the ground. The RPG shadow would not be on the front of the man's clothing. It would be on the ground given that the light is coming almost directly from his right, therefore with the RPG protruding at 90 degrees from his front, this shadow would show up on the ground as such.
I also agree that his left arm is down with his left hand holding his crutch. The clothing maybe loose, but the sleeves are typically not baggy enough to be in the position shown if his left arm is supporting an RPG.
But, to me at least, the absence of the exhaust flash not being reflected on his back is the big give-away to me. If you have a different take on why there is no exhaust flash reflected on the man's back, I'd be open to open to it's consideration.
Now I know EXIF data can be hacked but once Photoshop has been introduced, I can't trust what little EXIF data remains behind.
I don't discount any of your analysis.
For the record I spent 1 year most recently at Casio in marketing and 13 years at Nikon, most recently launching the D3100 and D7000. There are certain telltale markings in any photo.
Michael
http://news.za.msn.com/gallery.aspx?cp-documentid=159380193&page=10
Love the blog btw, just starting following.