Intel doctored Al-Quada videos/images (to some anyway)

[Rahne Everson]

To avoid putting "TV fakery" in the title.

Anyhoo, some people have quoted a recent Paul Watson article about computer researcher Neal Krawetz, who's apparently written a program that can point out the error rates in a JPEG resulting from the lossy compression. Weeding out all the paranoid nonsense and getting to the actual claims, Krawetz did this to stills from Al-Quada videos and found that parts of the image had different error rates when compared to the original image, including the Al-Sahab and IntelCenter logos and Ayman al-Zawahiri in this image. He's said (though he's corrected himself) that the Al-Sahab and IntelCenter logos were added at the same time because they had the same error rates and that al-Zawahiri was added into the frame.

You can read more about it on this blog at Wired and here's the source code for the program for the more tech savvy people here.

I'm not that technically savvy, but I'm familiar enough with data compression to know this is dodgy at best. Okay, they're added, so what? How does this prove that the intelligence agencies and Al-Quada are working together? As the guy from IntelCenter pointed out in the Wired blog:

Ben Venzke of IntelCenter says his organization didn't add the As-Sahab logo. He points out that just because the error levels are the same for two items in an image, that doesn't prove they were added at the same time, only that the compression was the same for both items when they were added.

And even according to Krawetz himself:

I was finally able to reach Neal Krawetz at the BlackHat conference to respond to the questions about the IntelCenter and As-Sahab logos (Krawetz doesn't have a cell phone on him so finding him at the conference took a while). He now says that the error levels on the IntelCenter and As-Sahab logos are different and that the IntelCenter logo was added after the As-Sahab logo. However, in a taped interview I conducted with him after his presentation, he said the logos were the same error levels and that this indicated they were added at the same time. Additionally, after I'd written the first blog entry about his presentation, I asked him to read it to make sure everything was correct. He did so while sitting next to me and said it was all correct. He apologizes now for the error and the confusion it caused.

Some of the comments are good info sources too:

Maybe it's just me, but I do not see how this will work for videos.

If the video was shot in front of a "black sheet" (BTW - black?? chroma keys are green or blue) and the image was added during the editing I don't see how this would work, since there is no jpeg involved in all this. You would be working on the video in a high res format, and would later compress it to the final product.



If the video had been edited with less advanced technology, you would probably not need any special analysis at all to know if it had been chroma keyed.


Also consider that the final mastering of the video on (from what we hear on the news) VHS tapes or files delivered through the web will degrade the image so badly that any analysis on this highly compressed material (in the VHS case, this will have even undergone DA/AD conversions) would appear to be completely pointless.


If only the still had been modified, again this analysis would be pointless as you would only need to compare it to the original video to know the difference.


Besides, the "error level subtraction" method does not make much sense to me. If you take a JPG image and save it at 90% quality, re-open it and re-save it at 90% quality, you will not have the same file the second time. You will be saving a 90% quality version of something that is already 90% of the original. So if you subtract the second image from the first, you will get a different error level all around the edges - where jpg compression is more visible. That appears to be exactly what is happening in those images, where areas that have more edges, and therefore present higher jpeg compression artifacts seem to stand out more. Notice the beard in the first image, the wraps of the robe in the second one - all edges always stand out. I do realize that this will work if you add a perfectly uncompressed image on top of a highly compressed jpg, as the resulting image will have a much lower error level in the freshly added images than in the original one, but this would be quite obvious to the naked eye as well. In case of similarly compressed images, I don't think this can be used to prove anything.


That being said, I might have a different opinion if I had been able to see this presentation in person.

Apparently this guy doesn't understand how DCT compression works. Or he's hoping that no one else does, and that they'll think he's a genius.


Not only is what he proposes to do impossible (unless he has access to the _uncompressed_ video master, he cannot distinguish source quantization from final video quantization), but what his "custom software" is doing is simply increasing the brightness in areas with high DCT coefficients for high frequencies (which happen wherever there is high contrast, fine detail or pixel noise).


In other words, he's basically written an "edge detect" filter that operates on the DCT tables. It's what compression experts describe as "DUH!"


BTW, the first image in this article is a chroma-key. You don't need any "custom software" to figure that out; just look at the blue spill on his beard. The _entire_ background is fake. It was probably taped in some cave, with a blue background, and then they decided to put him in a "library" to make him look more intellectual.

And one particularly long analysis by "Daisy Cutter"

If you watch the news you will see presenters, reporters and guests 'cut and pasted' against a backdrop such as the city skyline or the newsroom. This is not a new technique, it is a craft that has been attempted in every television studio, usually to get best results on those occasions when the people in front of the camera are not wearing whatever colour the back-drop is. Blue works best for skin tones, however green is a common choice as this does not conflict with jeans, blue shirts and other commonly worn items. Sometimes excess light can be picked up by other colours to make them blue/green, i.e. to be 'keyed out'. Normally a blue-screen engineer will work with the lighting 'spark' to get best results. The person presented as 'Mr Al-Q*eda' is wearing clothes that are easy to key, and with the camera correctly 'racked' there need not be any colour cast, and with a good engineer the 'fine hair detail' can be kept in, making the end results natural in appearance.
Regarding the back-ground, this always has to be a forgery of sorts. Small artistic decisions are needed to frame the subject. Skyline landmarks can be re-jigged to fit the frame and not conflict with the foreground. Because there is a foreground, i.e. the 'talent', interviewee or supposed propagandist, it is not necessary to have a complete background - why paint the floor if it is not going to be seen?
If the blue-screen route has been taken, then why spend too much time on a 'real' background? If you had to include ornate script in the background, would you paint it for real or just tweak a stolen image in Photoshop?
The point is that this promotional material for 'al-qaeda-ology' is no different to mainstream news in that the foreground is pasted against a background - standard production techniques.
Personally I don't find the photoshop quantisation error stuff particularly convincing. I have my own eyes and I can see the studio lights glaring on the forehead of the subject. These would have been necessary to light the subject if using Bluescreen, potentially using up to 10K watts of tungsten power, if properly key-lighting, back-lighting and fill-lighting the subject. With 'pretty' subjects a make-up artist can apply powder to control the shine of the hot lights on the forehead. The subject could have been shot against black, but he is wearing black, so blue/green would be easier, and likely to produce quicker results.
What happens to all of this light? We can all remember physics and how 'light gets darker' the further away you are from it. In the background scene there appears to be a desk, however not very much light has fallen on it. It is lit differently to the foreground, and therefore could not have been in the same space at the same time. Of course this could have been an al-qaeda ruse, maybe they wanted to out-wit the intelligence experts by using a telephoto lens and an odd angle on the subject that somehow keeps the backdrop verticals exactly parallel (not diverging).
The logo in the corner could have been forged by 'al-qaeda' however they could have also used standard production techniques. The 'bug' has an alpha channel and simply gets 'keyed' over, normally by a dedicated box, or just with an extra layer with 'Avid' editing tools.
If the person that spent so much time and effort on this study with normal news footage studio shots then something might be learned - the pictures are regular TV. As it is there is no 'control' presented and only guesses can be hazarded as to what is supposed to be 'proved'.
There you have it - almost. The noise levels of the image components will be different in composited images. The background is a computer manipulated still image (or stream of moving images). This might have no noise if it is crudely rendered in CG or altered significantly in Photoshop. Meanwhile, the foreground - from a normal studio camera - will have artifacts of noise introduced by the CCD and any tape medium used to store the pictures until the editing suite is reached. If using an analog studio - as is common - the camera heads and mixing desk will add their own noise - 'signal degradation'. When the foreground is matted to the background a mix of noise levels will be in the scene. Other interesting artifacts will result when the resultant image is put back onto tape. In the background there are areas that look great on a computer screen, but cause issues for the NTSC/PAL colour space. Not all colours will be quite the same and huge colour/contrast gradients will have to be flattened out. The original image could have one completely black pixel next to one completely white pixel, or a completely saturated red pixel next to a green one - such messing with sine waves is not do-able with 'NTSC/PAL' so more noise gets added, as appears to be the case here.
This article proves that people with access to a news studio produced this video - I am glad I know nothing new.
Where was I?

Some though, are very sad:

Was Krawetz threatened? Is that why he is reversing his story now? Was his family threatened.


Keep an eye on this guy and see if he doesn't meet with an accident in the next few months.

And to Daisy Cutter:

Daisy Cutter is a disinformation pig.

Oh boy...

 

[e^n]

In other words, he's basically written an "edge detect" filter that operates on the DCT tables. It's what compression experts describe as "DUH!"

Pretty much exactly, I posted in a funny thread about this over at LC.

http://z10.invisionfree.com/Loose_Change_Forum/index.php?showtopic=13124

 

[T.A.M.]

Don't let him meet ACE or he'll be Bukakked with stupid...lol

TAM

 

[uk_dave]

e^n,

I don't know if you're a masochist (or maybe even a sadist ) but I have to admire your tenacity in actually arguing with the kids at LCF.

BTW: Loved the question about chemtrails

 

[e^n]

I don't know if you're a masochist (or maybe even a sadist ) but I have to admire your tenacity in actually arguing with the kids at LCF.

BTW: Loved the question about chemtrails

I'm bored, plus it's hard to get me angry