Join 3,559 readers in helping fund MetaFilter (Hide)


Victoria's secret, uncovered by science
January 6, 2010 8:51 PM   Subscribe

Body by Victoria. It started with an invisible handbag. Photoshop Disasters mocked Victoria's Secret for running a shot of a dress model clutching the straps of a digitally wiped-out purse. Then Neal Krawetz at Hacker Factor got into the act, analyzing the image to show that the photo editors had not only swiped the model's purse, they'd toned her arms, enlarged her breasts, and lightened her skin. In the comments, and in this follow-up post, tons of excellent nerdery about how to tell the photofaked from the real, by science.
posted by escabeche (97 comments total) 58 users marked this as a favorite

 
I'm impressed and educated after that. I also want to propose we start using "pixelplating" to describe these things.
posted by Iosephus at 8:58 PM on January 6, 2010


Seeing is no longer believing. Which should have interesting consequences for criminal law.
posted by Chocolate Pickle at 9:03 PM on January 6, 2010


in Oz at the moment at there's a brouhaha here regarding photoshopping too. Nude image of Jennifer Hawkins on a Marie Claire cover is justified !

Although smacks of Pepsi Blue as Channel 7 here in Oz are closely tied to Pacific Mags
posted by doogyrev at 9:09 PM on January 6, 2010


I'm interested by the methods used for analyzing these images, but at least in this presentation it feels rather unscientific. There's no mention of false positives (except in one of the comments), and there is only some vague sketches as to why these "tests" make theoretical sense. The way they use PCA is the most worrisome part; I don't see why "outliers" in the color space are necessarily the result of digital modification. It's clearly not my field, so I can't say it's wrong, but the link doesn't exactly convince me that the methods work. Fun stuff though, and certainly worth a careful look.
posted by kiltedtaco at 9:20 PM on January 6, 2010 [3 favorites]


Excellent and interesting post. I am in partial agreement with kiltedtaco; I think their methods are not evidence and they should have used 'softer' language to indicate this. Still, fascinating and interesting analysis.
posted by Bovine Love at 9:25 PM on January 6, 2010


Seeing is no longer believing.

This is fine art retouching compared to the darkroom days, but it's still nothing new. The old print library at my last newspaper was hilarious: lots of thickly applied white-grey paint to touch up or remove people/buildings/whatever. It reproduced fine in print once it had gone through a coarse line-screening and rotten printing press, of course.

As Stalin knew, the camera has always lied. Now it's just smoother.
posted by bonaldi at 9:27 PM on January 6, 2010 [8 favorites]


I saw this a few days ago on HN; I'm glad somebody put together a good FPP on it. The techniques used are really fascinating, and even though I dabble in digital photography I was surprised at the amount of information he was able to extract. Most of it is a consequence of how JPEG works; if the photographer's workflow had been pure RAW and then output to PNG, without ever hitting JPEG, some of the techniques would not have worked (at least not in the same way, or as well).

I only wish there had been a bit more information on the tools and procedures used; every once in a while I've come across things that just seem so highly improbable, but nonetheless aren't obviously fake, that I'd really like to do something similar with them.
posted by Kadin2048 at 9:42 PM on January 6, 2010 [1 favorite]


Wow, how cool that they're able to reverse-engineer an image like that. I'd love to learn to do that.

Why do people continue to think that they can get away with making such huge alterations? I just don't get it. Even were it not for the amazing techniques featured here which sussed out the other work done, surely someone would have noted that, in all other pictures, this model is darker-skinned and smaller-breasted than shown in that particular image.
posted by po at 9:48 PM on January 6, 2010


Why do people continue to think that they can get away with making such huge alterations?

Because they can get away with it. Not one person in a thousand knows or cares to know how to do this kind of analysis. And they also don't care if the pictures they see in catalogs have been manipulated in this way.
posted by Chocolate Pickle at 9:55 PM on January 6, 2010


here is a paper (Note: it's a very large PDF and the web server doesn't tell you how large it is, so you might want to right click and save it's 5.84 mb). The author seems to think it's a 'magic bullet' for finding photoshoping but that's hardly the case. I'll explain more at the end of this post, but if you look at the girl's dress, you can see that there's a tight knit pattern that would not compress well in a JPEG, which could explain why you see 'spots' in the ELA.

in fact, the area with the missing bag doesn't appear at all in the "Error level analysis" proving that the known false negative rate for this method on on this image is 100%.

My conclusion is that this guy is a complete idiot pedaling a kind of visual snake oil, looking at random process effects on an image and then making authoritative statements when in fact he's really just making stuff up.

I mean look at this:
Although the dress appears to have a random noise pattern, there is actually one area where there is a well-defined pattern: her chest. Between her breast the dots form a well-organized "stretch" pattern.
MAYBE THE PATTERN OF THE DRESS LOOKS 'STRETCHED' BECAUSE THE DRESS ITSELF WAS STRETCHED, DUMBASS.

Seriously there is nothing "scientific" about this analysis. It's all bullshit.

---

ELA works by reapplying JPG compression over and over again on an image. As you know, JPG is a glossy compression method which slightly changes the way an image looks. But ELA won't work properly on an image that's only been saved once or twice. In fact on page 17 you have an ELA analysis done an an image that was not photoshoped to illustrate how 'fresh' images still have high ELA levels. And more importantly it only works if the original source image was a JPG, which hasn't been resized the 8x8 block nature of JPEGs is key for this thing to work. In other words it's great for detecting photo shops that use random images off the internet, but it's very unlikely that the artist didn't have the original raw (or very large) source images from the photo-shoot to work with)

How it works is kind of interesting but I don't have time to go into all of it. But basically like I said, this isn't "scientific" it's just random nonsense from some idiot who doesn't know what he's talking about.
posted by delmoi at 9:56 PM on January 6, 2010 [46 favorites]


Wow, how cool that they're able to reverse-engineer an image like that.

He wasn't. He's just spouting bullshit.
posted by delmoi at 9:57 PM on January 6, 2010


I'm going to assume that what happened here is that the photo was manipulated originally for print purposes, in which case the handbag would have been more ideal.

For shopping cart purposes, it was then given to some 17-year-old lowly, underpaid web designer who worked until their little heart burst trying to master the clone tool once and for all before realizing it was 4am.

Completely forgetting to work on the hand before uploading the image, she high-fived the server while jumping out the door realizing Starbucks [insert your favorite local/too-cool-for-school coffee joint] was open and her Australian raiding guild was just about to get started.

.. shit.
posted by june made him a gemini at 9:57 PM on January 6, 2010 [4 favorites]


The step-by-step illustration of the techniques used is really fascinating, but holy hell are some of those negative example images frightening.
posted by cmgonzalez at 9:58 PM on January 6, 2010


I am a graduate student in computer vision. Having briefly looked over his paper, the basic idea (that compositing parts of two different jpg images [each with their own quantization matrices] together should show a discrepancy in artifacting) is sound, but his actual techniques seem rather ad-hoc, and he presents only qualitative results, which is suspicious. His use of PCA is questionable.

On preview, what delmoi said, but less harsh.
posted by Pyry at 10:01 PM on January 6, 2010 [5 favorites]


Also thank you delmoi, I'd made a four-paragraph rant about how this was all crap and any hodgepodge of filters will achieve the same effects without the ironic bullshit of bullshit and then decided that I'd take another route.
posted by june made him a gemini at 10:02 PM on January 6, 2010


.. also that recoloring "fiasco" is complete poppycock. She was bronzed to all get-out on that runway.
posted by june made him a gemini at 10:05 PM on January 6, 2010 [1 favorite]


Most of it is a consequence of how JPEG works; if the photographer's workflow had been pure RAW and then output to PNG, without ever hitting JPEG, some of the techniques would not have worked (at least not in the same way, or as well).

Look at the images where we know the handbag was removed. There is no difference at all, nothing shows up in the ELA.

Not only does the ELA require you to use JPG and re-save multiple times you can't change the size of the image as you work on it otherwise the 8x8 blocks won't match up!

Do you seriously think the artist worked on the image at that size and in JPG at the same time? Of course not! And that's why the removed bag doesn't show up in the ELA!

The difference in "noise" could easily be due, again, to the tight pattern or texture of her dress.

And actually I think it's possible that they enlarged her breasts. But the "technique" he used to find the modification could easily have been caused by the underlying fabric being stretched, like I said.

But come on, like I said, there is no way the artist used images these size, rather then the originals when doing this work, which means the ELA wouldn't work at all. And if you compare with the actual examples from the paper, the modified images have huge regions that don't align with any objects in the image itself.
posted by delmoi at 10:09 PM on January 6, 2010 [1 favorite]


Most of y'all probably know about this fun and entertaining blog already, but for those that don't... Photoshop Disasters.
posted by flapjax at midnite at 10:09 PM on January 6, 2010 [2 favorites]


I've never seen the word "nerdery" in the same paragraph with "Victoria's Secret" before.
posted by twoleftfeet at 10:10 PM on January 6, 2010 [1 favorite]


.. also that recoloring "fiasco" is complete poppycock. She was bronzed to all get-out on that runway.

Yeah, people look different under different lighting/camera settings/etc.
posted by delmoi at 10:11 PM on January 6, 2010


Ouch. Even if the techniques used to ascertain certain flaws are questionable and the blogger is spewing utter bullshit, this is still a big Photoshop flop.

I'm beginning to wonder why anyone bothers to enforce rigorous standards on models anymore; When a company can modify every aspect of a model in Photoshop, why bother choosing a specific model? Just create from your back catalogue!
posted by far from gormless at 10:16 PM on January 6, 2010


I'm a graduate student in computer visualization, and image processing isn't my specialty, but I agree with what Pyry said. There are some interesting ideas here, but all the evidence is qualitative. I think that a paper using the ELA approach and more rigorously applying it to (say) 50 photoshopped images and 50 non-photoshopped images to get some baseline information would be very interesting, but there doesn't seem to be one here. Applying the techniques to mostly unknown datasets doesn't give me any confidence.

Also, remind me never to put Dr. in front of my name whenever I get my Ph.D. in CS. For some reason, it makes you sound like you don't know what you're doing.
posted by demiurge at 10:19 PM on January 6, 2010 [1 favorite]


JPG is a glossy compression method

Which is why this sucks; everybody knows real photoshoppers use matte compression methods.
posted by Monday, stony Monday at 10:26 PM on January 6, 2010 [8 favorites]


(but indeed, the article really sounds like someone making shit up)
posted by Monday, stony Monday at 10:27 PM on January 6, 2010


JPG is a glossy compression method

I prefer a nice matte finish GIF, myself.
posted by abc123xyzinfinity at 10:28 PM on January 6, 2010 [2 favorites]


But seriously,

MAYBE THE PATTERN OF THE DRESS LOOKS 'STRETCHED' BECAUSE THE DRESS ITSELF WAS STRETCHED, DUMBASS.

The image noise is created on capture and randomly distributed; it doesn't correspond to any patterns in the subject matter, so the noise wouldn't mimic the shape of the dress. However, if the artist pushes pixels around in photoshop, he/she fucks with the distribution of noise, hence patterns in the noise indicate digital manipulation. That seems pretty reasonable to me.
posted by abc123xyzinfinity at 10:34 PM on January 6, 2010 [4 favorites]


How can you not look at that original picture and not realize the teeth and dress aren't modified? And once that happens, isn't the rest just obvious? Or at least...well...questionable?

I'm not saying this post isn't interesting (it is) and that the content (even if it's questionable) isn't worthy of an FPP, but if you don't realize the nearly EVERY. DARN. PICTURE. you see that is meant to sell you something isn't doing just what's been outlined here, you need to pay closer attention.

Seriously.
posted by MCMikeNamara at 10:38 PM on January 6, 2010


I'm glad I wasn't the only one who suspected I was being bs'd about halfway through reading that.
posted by empath at 10:38 PM on January 6, 2010


Um, what is up with that?
posted by gottabefunky at 10:38 PM on January 6, 2010


However, if the artist pushes pixels around in photoshop, he/she fucks with the distribution of noise, hence patterns in the noise indicate digital manipulation.

Assertions aren't allowed to prove themselves, y'know.
posted by Mikey-San at 10:45 PM on January 6, 2010


(i got no dog in this race, i just felt that was worth pointing out)
posted by Mikey-San at 10:45 PM on January 6, 2010


However, if the artist pushes pixels around in photoshop, he/she fucks with the distribution of noise, hence patterns in the noise indicate digital manipulation.

Sure, but the problem is that there isn't a way (in general) to reliably distinguish between noise and legitimate high frequency information (if there were, we could perfectly 'denoise' images, instead of using the glorified blurs that are available now).
posted by Pyry at 10:45 PM on January 6, 2010 [3 favorites]


Um, what is up with that?

Leonidas will jump out of the woods and save them any moment now.
posted by ROU_Xenophobe at 10:48 PM on January 6, 2010


JPG is a glossy compression method

er, lossy. Damn firefox spellchecker.
posted by delmoi at 10:52 PM on January 6, 2010


Victoria's Secret put the handbag back, it looks like?
posted by anazgnos at 10:59 PM on January 6, 2010


The image noise is created on capture and randomly distributed; it doesn't correspond to any patterns in the subject matter

Of course it does.

Well, it depends on what you mean by 'noise'. Obviously if you take a picture of something noisy when you analyze the picture the noisy object is going to show up as having more 'noise'. And that's exactly what the dress was. It had a texture with features smaller then a pixel, and it would have a lot of 'noise' at that resolution.

Noise detection algorithms have no way to determine if noise is "real" or if it actually exists in the image, which is why you see people talk about losing detail to noise reduction algorithms on sites like dpreview.com
posted by delmoi at 11:04 PM on January 6, 2010


I'm about 100% positive the author of the hackerfactor link ran the image though a bunch of filters and then fabricated a back-story.

In other words, y'all postin' in a troll thread.
posted by lekvar at 11:13 PM on January 6, 2010


Ok, to formalize his idea in a bit more principled way, in the following situation it should be possible to tell which 8x8 blocks have been altered:

1) A compressed jpg image is used as the base
2) Some alterations are made without scaling or cropping the image (i.e., only compositing stuff on top of the base image)
3) A significant portion of the base image is unchanged
4) The resulting image is saved at a higher quality level than the base image

Basically, jpeg divides the image into 8x8 blocks which are each transformed by the discrete cosine transform into a matrix of coefficients, which is then quantized.

So if we were to flatten the matrix into just a long row vector, an 8x8 block might be represented by a bunch of coefficients like [8, -6, 12, 34, 1, ...], which might then be quantized (rounded down) to the nearest 5, to get a quantized vector like [5, -5, 10, 30, 0, ...].

Now, if you were to resave this block with a higher quality level (say rounding to the nearest 2 instead of 5), you would get [4, -4, 10, 30, 0]. If you were to look at the distribution of values over a bunch of blocks, you would see that the values 2, 6, 8, 12, 16, 18, 22, 26, 28, ... never occur, because there are no multiples of five that when rounded down to the nearest multiple of 2 produce these values. E.g., 5 -> 4, 10 -> 10, 15 -> 14, and so forth, so you can't produce a 2 by rounding to the nearest five and then the nearest two.

In short, by resaving with a higher quality level you will produce distinctive gaps in the distribution of values, which can potentially be identified. If you modify the image, then the 8x8 blocks which have been modified can have values in those gaps, especially if you modify the image by directly painting on it rather than pasting in another image (pasting in another image will only avoid gap violations if you align the pasting with the 8x8 blocks, and the pasted image used exactly the same quantization as the base). If a block has values in a gap, then it has to have been modified.
posted by Pyry at 11:17 PM on January 6, 2010 [15 favorites]


Assertions aren't allowed to prove themselves, y'know.

I don't know, I thought my assertion was about how digital cameras capture images and how photoshop alters them (at least with the kind of tools that make breast bigger). Both these points seem fairly incontestable to me.

Sure, but the problem is that there isn't a way (in general) to reliably distinguish between noise and legitimate high frequency information

I may be misunderstanding you, but it seems that there's no need to solve this problem "in general" in this context. We're talking about a specific image in which there is a way to distinguish the true noise from the alterations, because we have the additional information of knowing what we're looking at: a woman's breasts. If the question is, "Were these breasts digitally manipulated," the evidence presented seems to suggest yes, with a greater degree of confidence than simply comparing another photo side by side or falling back on the fact that all these images are manipulated in one way or another.

This technique doesn't have to work on every image all the time, but it works here.

And that's exactly what the dress was. It had a texture with features smaller then a pixel, and it would have a lot of 'noise' at that resolution.

Sure, you're right. The dress overall is noisier than other parts of the image, but you wouldn't expect much variation in noise across the dress.

Noise detection algorithms have no way to determine if noise is "real" or if it actually exists in the image

Right, so my understanding is that the whole reason this process works is that the computer does not make the final decision as to what is noise and what is not, the human does. A human might not be able to distinguish noise from signal reliably for any old random image, but this is not a randomly selected image. Given the narrow search criteria, a false positive seems that much less likely. Is that not fair?
posted by abc123xyzinfinity at 11:18 PM on January 6, 2010


the whole reason this process works is that the computer does not make the final decision as to what is noise and what is not, the human does.

I bet there are a bunch of humans who can directly look at an image and reasonably reliably tell you whether it has been altered, but that's not science.
posted by Pyry at 11:25 PM on January 6, 2010 [1 favorite]


The image noise is created on capture and randomly distributed; it doesn't correspond to any patterns in the subject matter, so the noise wouldn't mimic the shape of the dress. However, if the artist pushes pixels around in photoshop, he/she fucks with the distribution of noise, hence patterns in the noise indicate digital manipulation. That seems pretty reasonable to me.

I'm not sure about that, if the manipulation takes place at a higher resolution, and then is scaled down, it won't have much of an impact. And anyone not doing a fark photoshop contest in five minutes is going to do their work in high resolution, and work on uncompressed images.

Like I said, I do kind of think they did some work on the breasts. But I'm not going to claim certainty like the author did.

But look at the dress, it's form fitting, that means it's naturally going get stretched itself by the body, So even a natural, unmodified image could still have those artifacts.
posted by delmoi at 11:30 PM on January 6, 2010


I bet there are a bunch of humans who can directly look at an image and reasonably reliably tell you whether it has been altered, but that's not science.
--
Like I said, I do kind of think they did some work on the breasts. But I'm not going to claim certainty like the author did.

Fair enough. For what it's worth, you can color me more skeptical than when I came to the thread. I guess I've been watching too much CSI...
posted by abc123xyzinfinity at 11:36 PM on January 6, 2010


That wolf photo looks like a wolf's head pasted onto a dog's body. There's a pretty obvious triangle on top of the head. You can see the outline of the bad cut out if you brighten it enough. My guess is the dog is a Portuguese water dog. Also the feet do not look like wolf feet. They look like poodle-type feet, like those of Portuguese water dogs.
posted by wherever, whatever at 11:38 PM on January 6, 2010 [3 favorites]


Where is that wolf photo from?
posted by delmoi at 11:48 PM on January 6, 2010 [1 favorite]


Doing the work in pshop is always the fallback when things don't work on the shoot. So much easier to just get it done as the photographs are being taken, but there's always a time restriction (losing light, etc) that forces it into digital manipulation. Once in the studio, it's never about reality; it's about creation of aspirational images. When you're able to change anything the shot itself isn't real anymore - it feels more like a game. It's hard to describe, but there is a disconnect. And of course, it's about selling whatever crap is being produced.

After spending years working on the production side in ad agency and design firm studios, I showed a couple of teacher pals how to spot the work done when going through those fashion mags that their teenage students love to read. Once you've been in it for years you don't need fancy ELA to know what's gone on, it becomes instinctive, you see perfection where you know it shouldn't exist. (And yeah, I'm with delmoi on a lot of that 'scientific' analysis.)

I think for all the discussion of "media awareness" that people tout, the mind can only fight for so long against a constant onslaught of computer-generated beauty. The part that is sad is that so many models I worked with were just so beautiful and showed such personality, then it'd get filtered away. I often thought people would relate more to a shot where the only work we'd done was lighting corrections & very slight colour work to compensate for the flatness of the shot.
posted by Salmonberry at 12:24 AM on January 7, 2010 [3 favorites]


Boobs altered during TV broadcast.

Honourable mention - source unknown.

[not gratuitious, no nudity, but probably NSFW]

posted by uncanny hengeman at 12:36 AM on January 7, 2010 [9 favorites]


Ok, it looks like the wolf photo is explained here.
posted by sentient at 12:44 AM on January 7, 2010 [3 favorites]


I swear that's a hyena's body in the "wolf" image.
posted by maxwelton at 12:56 AM on January 7, 2010


Man that leg thing is completely noticeable now.
posted by june made him a gemini at 12:58 AM on January 7, 2010


*zooms in on boobs, presses 'enhance'*
posted by obiwanwasabi at 1:11 AM on January 7, 2010


Actually, it's a dog running away from the camera. All I did was paste a black circle over the "head".
posted by maxwelton at 1:15 AM on January 7, 2010 [1 favorite]


I don't have any doubt whatsoever that nearly everything in this shot is faked: boobs, smile, skin tone, handbag-thing, hips, dress. just obvious to me as an art director. only what I don't get is why this is such an outrage. it's an ad, of course it's an attempt to show their product in the best possible light. of course those people are insecure enough not to just depict reality but a perfect fantasy. does that really surprise anyone?
posted by krautland at 1:27 AM on January 7, 2010 [1 favorite]


it's an ad, of course it's an attempt to show their product in the best possible light. of course those people are insecure enough not to just depict reality but a perfect fantasy. does that really surprise anyone?

What it is is another illustration of the fact that, if many people think they can cheat undetectably, they will cheat and cheat and cheat and cheat.
posted by JHarris at 2:04 AM on January 7, 2010


Who's outraged? Or surprised? It's a how-to article, showing people how to detect the manipulation if they're not practiced at it.
posted by harriet vane at 2:41 AM on January 7, 2010


Is this really still up for debate? Shock horror, Photoshop is used to make someone look pretty. BoingBoing is still dining out on this revelation too.
posted by monkeyJuice at 2:41 AM on January 7, 2010 [1 favorite]


The dress overall is noisier than other parts of the image, but you wouldn't expect much variation in noise across the dress

You're still talking about "noise" as if it's an actual distinct thing, and ignoring that all the algorithm is detecting is some mathematical feature of the image's numerical data (though the description is too vague to know exactly what) that allegedly corresponds in some way to where noise might be.

The algorithm has no way to tell actual noise from image features that have similar mathematical properties to noise, and so delmoi's entirely right to say we can't tell whether the noise has been stretched or if the stretching of the dress is causing a pattern to show up that's mathematically indistinguishable from stretched noise.

Though if you're impressed by this guy's analysis, just wait until you see what the scienticians in this video can do.
posted by cillit bang at 2:52 AM on January 7, 2010 [3 favorites]


Wow, how cool that they're able to reverse-engineer an image like that.

He wasn't. He's just spouting bullshit.
posted by delmoi


I'm gullible sometimes, alas. I guess I just wish I could return to the late 90s when everything on the Internet was new and cool and I could conceivably master it all and figure out all of its tricks and be the most awesome at awesome-ing. Thanks for the rundown, delmoi.
posted by po at 2:53 AM on January 7, 2010


Background on my first post. I just got a horrible feeling the background story could all be BS, so I did a quick Google for more info.

Apparently, the producers of Dutch Home Improvement show "De Grote Woonwens" (free translation: The Big Wish of Living), thought female presenter Nicolette van Dam could really use some bigger breasts. Instead of using bra fillers, they chose to use computer re-touching of the footage. Sadly, they didn't really do a convincing job of it. In a clip from the program, which was first featured on satirical website "GeenStijl" and later uploaded to YouTube, the viewer sees the breasts of Nicolette van Dam grow spontaneously during the show.
posted by uncanny hengeman at 3:06 AM on January 7, 2010 [4 favorites]


tons of excellent nerdery about how to tell the photofaked from the real, by science.

hey, i can do that without science!

- photofaked: anything you see in magazines, advertisements, etc
- real: what you see everyday

i don't see what's so particularly difficult or complex about that.
posted by UbuRoivas at 3:07 AM on January 7, 2010 [3 favorites]


Neal's a friend of mine, we're both in the security space, and we're both graphics guys too.

He's not a fraud. He's just done a really poor job not looking like one. What? A geek having issues recognizing how other people will react to something? Never :)

Hopefully there will be better data soon. I'm pretty confident at least some of his techniques will hold up after all.
posted by effugas at 3:20 AM on January 7, 2010 [5 favorites]


uncanny--

That's the funniest thing I've seen today. Wow.
posted by effugas at 3:21 AM on January 7, 2010


The bag handle was pretty funny. What a hack job. Missing legs etc. can be hard to notice if you are not looking, although they are hard to ignore once noticed. The bag handle, though, that is precious.
posted by caddis at 3:25 AM on January 7, 2010


Seriously there is nothing "scientific" about this analysis. It's all bullshit.
I'm not an expert in image forensics, but it is a real field. He did a pretty impressive presentation at Black Hat and I know at least one other group has written similar software to his using the algorithms he described at Black Hat.

There was an interesting back and forth between Neal and Dan Kaminsky on funsec. I'm not at all sure that Neal's tools do everything he says that they do, but he talks the right language in the right places. It isn't entirely BS and snake oil, although nor it is fully accepted by those who know. The DC3 are, indeed, the best in the world at this kind of thing. If they certify his tool, it is real.
posted by Lame_username at 4:36 AM on January 7, 2010


Oh wow... Look at the eyes of the guy standing next to the girl at 0:16 of the embedded video. He appears to check out her boobs (so funny to catch guys on cam doing this), looks again as they change size (coincidence, I'm sure), and then looks up and away like "I see NOTHING!!"
posted by LordSludge at 4:46 AM on January 7, 2010


Not to take sides or pretend to have knowledge I don't, but I hope to God there's a delmoi in every courtroom where stuff like this is presented as evidence.
posted by rahnefan at 4:51 AM on January 7, 2010


That's a lot of work to not really prove something that doesn't really matter anyway.
posted by fire&wings at 5:03 AM on January 7, 2010 [1 favorite]


I feel like we owe that guy who could tell if a pic was shopped by its pixels an apology...
posted by samsara at 5:31 AM on January 7, 2010


delmois: In other words it's great for detecting photo shops that use random images off the internet

I don't know why you have to go so over the top and follow this up with the "it's all bullshit" stuff.
posted by smackfu at 6:06 AM on January 7, 2010


First, Props to Delmoi for nailing all of this early on.

One thing that rings false to me is the assumption that the retouching was done onto a compressed jpg in the first place. This goes against most agency workflow I've ever seen. (yes, I used to be a retoucher in a large regional ad agency) Normally, any photo work is done on a large original which can be in any format but is then saved as a psd or tiff with the work usually on a separate layer. You then would save out a jpg and size the image for web use and optimize. This would eliminate most of the "evidence" listed in this article, because it all flows from the assumption the retouching leaves artifacts in the jpeg. The jpeg is created after the fact as a web-specific file when the work is finished.
posted by Mcable at 6:31 AM on January 7, 2010 [1 favorite]


...a complete idiot pedaling a kind of visual snake oil...

Boy, they don't make snake oil like they used to. For one thing, in the old days, it didn't have pedals. The best stuff was invisible, of course.
posted by Kirth Gerson at 6:55 AM on January 7, 2010


delmoi : But basically like I said, this isn't "scientific" it's just random nonsense from some idiot who doesn't know what he's talking about.

I will entirely agree that his analysis involves a lot of subjective evaluation, thus it lacks proper scientific rigor by definition. Comparing it to the re-released and less-processed image, however, pretty much gives him a 10/10 for getting the right answers. So you really can't justify saying "He's just spouting bullshit" without calling him either psychic or an incredibly lucky guesser.

That said, it does really bother me that such a subjective analysis could very well help send someone to prison in a different context.

And as an aside - Personally, I think the less-processed version looks quite a lot nicer. The so-called "artist" didn't make her look better, he made her look like Tammy Faye.
posted by pla at 7:01 AM on January 7, 2010


this isn't "scientific" it's just random nonsense from some idiot who doesn't know what he's talking about.

The best way to respond to over-the-top claims is not with your own over-the-top judgments. A bit of caution on both your parts would have saved a lot of beanplating.

ELA has its place in image forensics. Many of the technical objections in this thread are valid, but these objections do not invalidate the field at large. You'll find a lot more interesting techniques coming from Hany Farid's lab. I've seen him talk two or three times, and his approach is as measured as it should be, but still full of interesting and novel approaches to detecting image forgery. Before you make any judgments about the field, you should seek some of his lectures online. You'll see that he tailors specific techniques to specific images and types of forgeries, and that he also advocates for a carefully chosen combination of approaches to "prove" any particular forgery.
posted by fake at 7:11 AM on January 7, 2010


Is it just me or is there not much of a significant practical difference between the "new image" that supposedly proves all his assertions, and the "old image"? The way this guy writes, with that 100% certainty colouring everything, it's as if he uncovered some vast conspiracy theory to tweak images, but relative to some of the really egregious photoshop modifications I've seen (where the model's significantly changed and thinned, or had body hair removed), this seems fairly mild. If anything, this would be more interesting and impressive if he applied it to something that wasn't clearly Photoshopped in some manner -- even his other analyses on his site are done on completely obvious photo manipulations.
posted by Big Fat Tycoon at 7:19 AM on January 7, 2010


It might have all been BS, but I sure do feel nerdier after reading it.

I tend to assume that basically any photo I see of a famous actor/model/famous person has been photoshopped as all get out--the digital age has made me exceptionally cynical in this regard. The discussion here is making me wonder if maybe I'm being a little unfair in some instances?
posted by Go Banana at 7:28 AM on January 7, 2010


Comparing it to the re-released and less-processed image, however, pretty much gives him a 10/10 for getting the right answers. So you really can't justify saying "He's just spouting bullshit" without calling him either psychic or an incredibly lucky guesser.

Exactly which things does he get right that you couldn't guess about any fashion image and be right 99% of the time?
posted by cillit bang at 7:34 AM on January 7, 2010 [1 favorite]


This analysis is hilarious. It's like he ran every photoshop filter out there and wrote a paragraph about how it "proves" his theory.

"And here, adding 10% lens flare clearly shows how the feet have been elongated!!!1"
posted by odinsdream at 7:38 AM on January 7, 2010


My conclusion is that this guy is a complete idiot pedaling a kind of visual snake oil, looking at random process effects on an image and then making authoritative statements when in fact he's really just making stuff up.

Hate to say, but that was my first impression too. I could easily tell just by eye that much of what he *said* had been done clearly was... and more... yet his breakdown just felt a bit laughable. Just from looking at it I can tell you that the skin tones were changed, the knit on that dress around her legs was ridiculously blurred, and man, the mismatching of the tiles on the ground was just embarrassingly amateur work. And only a blind person would think it was ok to leave those straps in! WTF? If I did poorly detailed Photoshop work like that, I'd have been quickly returned to paying my bills as a waitress decades ago.

If you know how to use Photoshop really well, when you are done with an image nobody should know it was ever touched digitally *but you*. Your goal is to make your work totally invisible and natural, to the best of your ability. It's better and smarter to leave a bad image somewhat alone than present sloppy digital work to an audience.

That aside though, as I was reading that guy's critique I couldn't help but feel he was just making up stuff that he thought sounded good and impressive... kinda like when people discuss a vintage of wine using big words when the truth is they don't really know anything about wine and are counting on you not to know enough about wine to know that they don't know anything about wine. So along the way they may be saying some stuff that's kinda accurate, but it's all layered in wacky verbage they totally made up.

Now I could be wrong, I've been wrong before, but in this case I'm really thinkin' I'm not.
posted by miss lynnster at 7:42 AM on January 7, 2010


In case anyone was wondering, it was a picture in a catalog circular (and those Victoria's Secret creeps send us at least one per week as of late), not some kind of award winning iconic photo. I doubt anyone working for the catalog cares all that much about being exposed.
posted by Burhanistan at 7:48 AM on January 7, 2010


maxwelton: "Actually, it's a dog running away from the camera. All I did was paste a black circle over the "head"."

Nice job. That pretty much settles it; I didn't notice the paw pads until you pointed them out.
posted by Kadin2048 at 7:56 AM on January 7, 2010


The biggest hint it was mostly snake oil to me was when he was like "oh, mah toolz? naw bro, they are too 1337 for yous, and they secret. yeah, no way yous smartz enuff to use dis so no you can't see the source code or downloader or use the program i usesed becuz yous aint 1337 enuff liek me!!!!!!!11111"

That destroys any semblance of science. I mean, he says he used some special programs to do this, but won't actually back up his findings with the rest of the peer review process, which is to say, his experiment design and tools. As far as I'm concerned, this fellow has good eyes and likes to read too much into ugly photoshop filters. This sort of secret-magic analysis would never fly in my courtroom.
posted by fuq at 8:40 AM on January 7, 2010 [2 favorites]


Comparing it to the re-released and less-processed image, however, pretty much gives him a 10/10 for getting the right answers.

Really? Where's the darker woman we were led to expect from the bronzed runway photo? In the photo used for the circular, she actually looks marginally darker than the original. Suddenly his photo-comparison commentary wrt skin is all about contour and very little about colour.

Of course, it is one thing for me to accuse Victoria's Secret of recoloring the skin on their model. It is another to prove it. Here are two photos of the same model.

This word, "prove"...
posted by Durn Bronzefist at 9:02 AM on January 7, 2010


I will entirely agree that his analysis involves a lot of subjective evaluation, thus it lacks proper scientific rigor by definition. Comparing it to the re-released and less-processed image, however, pretty much gives him a 10/10 for getting the right answers. So you really can't justify saying "He's just spouting bullshit" without calling him either psychic or an incredibly lucky guesser. -- pla
What do you mean? I noticed they changed the colors a little bit on the new image, but that might just be a new artistic choice. It didn't look like they changed anything about the boobs or her arms or whatever (I mean, there is no difference in her boobs or arms from the old Photoshop to the new one on their site with the handbag restored)

And his "methods" didn't detect the missing handbag at all, there's nothing that shows up in ELA or any of his other filters.

Anyone can look at the image and say "hmm, kind of looks like they enhanced her boobs". The other stuff is highly speculative and the "I can tell from the (processed) pixels" thing is way off.

Being a "lucky guesser" and then making up all kinds of pseudoscientific nonsense to justify it is still bullshit, IMO. And we really don't know which of his guesses are true. The boobs, probably. Did they really tone her arms though? Did they change her skin tone outside of the bounds of normal adjustment?
I don't know why you have to go so over the top and follow this up with the "it's all bullshit" stuff. -- smackfu
erm...
ELA has its place in image forensics. Many of the technical objections in this thread are valid, but these objections do not invalidate the field at large. You'll find a lot more interesting techniques coming from Hany Farid's lab. I've seen him talk two or three times, and his approach is as measured as it should be, but still full of interesting and novel approaches to detecting image forgery. Before you make any judgments about the field, you should seek some of his lectures online. You'll see that he tailors specific techniques to specific images and types of forgeries, and that he also advocates for a carefully chosen combination of approaches to "prove" any particular forgery. -- fake
fake, I never said ELA didn't work. But I did read the paper I referenced. There are some important restrictions, including not changing the image size as you work on it. ELA (essentially) works by detecting how many times each 8x8 block has been saved as a JPG. So if an image has been saved over and over again, then when you add things they'll have a higher ELA level. If you read the paper they have an example of doing ELA on an image that had only been saved once. The whole image showed up as having a high level.

Tailoring the technique to the image, and what you know about it is a very important point. Working in high resolution and then scaling it down would defeat ELA, and a lot of the other techniques are just detecting the high image frequencies in her dress.

On the other hand if you do tailor the techniques to the image you are working on, you're going to lose objectivity. It becomes more of an art then a science. Some people can do it right and some people will do it wrong. Claiming that running the photos through some filters can prove this or that are, I think, not correct. You're more likely to run into false positives. Especially when you still have people interpreting the result image.
posted by delmoi at 9:14 AM on January 7, 2010 [1 favorite]


I hate to break it to this guy, but virtually all model shots are altered to some degree. If you want a fun game to play the next time you're standing in the checkout line look at all the magazine covers and count the women over 30 who magically don't have bags or wrinkles under their eyes (some of them don't even appear to have lower eyelids). Then look for any wrinkles or irregular bumps in their skin, all gone. Most women's faces are reduced to eyes, nostrils and mouth floating in an undisturbed sea of flat skin, it's a bit disturbing when you notice how unnatural it looks.
posted by doctor_negative at 9:14 AM on January 7, 2010


This is why I always photomanipulate the psd and THEN resize and save it as a jpg.

I defy you to apply some 1337 filt0rz and see where I airbrushed mah b00bZ!

What?
posted by darkstar at 9:51 AM on January 7, 2010


Seriously, though, what kills me about the guy's analysis is that it kind of strains at a fernseed but swallows an elephant. The most glaringly obvious, indisputable photomanipulation in the whole thing is the removal of the purse, right? Yet in virtually none of his analysis images is this obvious photoshoppery evident in the way he describes for all the other things.

I mean, look at it. That area of the image should be glaringly manipulated in all of his evidentiary images. Yet the tile beneath where the purse used to be isn't really dramatically different from any other section of tile.

Which kind of suggests to me that his methods - and the explanations he uses - really aren't as "aha, gotcha" definitive as he'd like folks to believe, at least as they are applied to this photo.

I also agree with the comments upthread that the boob enlargement is practically imperceptible from the before and after photos. The manipulations in this one seem pretty mild artistic adjustments. The guy really could have done a much better job by choosing a photo that illustrated his techniques better.
posted by darkstar at 10:09 AM on January 7, 2010 [2 favorites]


I think they just took the purse out because they didn't want to confuse shoppers since it wasn't for sale.
posted by Burhanistan at 10:12 AM on January 7, 2010


How can you not look at that original picture and not realize the teeth and dress aren't modified?

Lots of people aren't conversant enough with digital media-- or don't spend time examining images closely enough-- to be able to spot changes immediately. And these images are potent and powerful, and they are ubiquitous. My nieces don't automatically believe that all media pictures of women are photoshopped; all they see are beautiful women who look far better than they do. This is the issue, that erosion of confidence that occurs daily; the rebuke of the unnaturally beautiful.

It would make things much easier if retouchers were given credit in a tagline, just as a signal that modifications other than makeup and good lighting were involved.
posted by jokeefe at 10:13 AM on January 7, 2010 [1 favorite]


Anyone who's interested in the world of photo retouching should check out this New Yorker profile of Pascal Dangin, Master Manipulator.
posted by cowboy_sally at 11:14 AM on January 7, 2010


I'm just impressed that the person in the photo actually exists. I expected less.
posted by blue_beetle at 11:32 AM on January 7, 2010


Tempest-Teapot.
Seriously. How is this even an issue? Photo manipulation has been around since photography has been invented. These sorts of edits were being done way before computers made is so easy a caveman could do it. Back then, though, it was far more painstaking work, done by very skilled masters.

Back in college, my design class took a tour of the art and photo department at Playboy in Chicago. The level of retouching done was amazing, as was the quality and skill of the work.

About the only "scandal" I take from this (as well as the recent Demi Moore kefluffle) is just how crappy the work is, an how low the level of acceptability has gotten. Those little details...the remaining straps in the hand, the missing mortar lines in the bricks...would have been taken care of by most any retoucher I have known over the years. In my mind, it seems to raise the question of whether quality or attention to detail even matters anymore. I'm not sure it does. As more and more of this sort of work work gets farmed out to cheap assembly-line art factories in India, Russia, etc. this sort of inattention to detail is going to become the norm. It just seems sometimes like our culture is hell-bent on seeing just how cheap and crappy things can get before we finally care. I'm not hopeful about where that bottom is.
posted by Thorzdad at 12:03 PM on January 7, 2010 [2 favorites]


Sorry if I sound like a broken record, but it is from a damn catalog. There was no intention of permanence here, it was a rush to print like most catalogs are. Who cares if it's sloppy? There's nothing here.
posted by Burhanistan at 12:08 PM on January 7, 2010


For those who want more science in their discussions about digital imaging manipulation and ethics;
The Science of photography..
Science is really pretty, and often more comprehensible when it's in image form
Like this article onThe Mathematics of Geometric Photograph Manipulation.
However those who say 'but reality isn't real anymore'.. are right, and vigilance is needed...
Photo-manipulation is a serious issue in the arena of Scientific Misconduct.

Which is why the journal Nature and others have instituted strict sets of ethical guidelines for scientists who submit papers to them.
If a speck of dust on the film obscured data in the original photograph, another picture was taken. Slicing films to rearrange the order of samples, or to splice in a control group that was actually part of another gel, was not common because it took almost as much skill to do that as to rerun the experiment.
We should conspire to end the fetish of the perfect image. Let's all get a little more 'real'. Nature is happy to work with others to aid the promulgation of image standards that we can all live with. The responsibilities of the institutes that train students, of the investigators who use their labour, and of the journals that publish the data can be better defined. Finding ways to regain our trust in scientific images is a goal on which we can all agree.


The journal of Cell Biology (and related journals) also felt the same way... in a field where images are vital, for example, to expressing the outcomes of electrophoresis gels.. there is great opportunity for misleading conduct. And great need of honesty, in order to keep the strong reputation of this field.

Digital image manipulation ethics, a bibliography of some important papers
How can we trust Anything?? An image based history of the history of photoimage manipulation
and the ethics for photography, some really interesting thoughts
What is our only hope??

Standards baby
(Nature Publishing Group's standards for authors submitting images.)
...and in newsrooms I think they are likely so far past the point of caring what they show while our pretty pretty newshosts show up and speak their pre-writen, target market focus group written, doublethink laced scripts; each word specifically chosen to create a set of ideas to work in harmony with the short clips and stock images most rousing to our emotional sensibilities loop endlessly in little boxes to their left or right... that this article doesn't matter.
Is a solution Forensic digital imaging?
Or a strengthening of the culture of academic honesty. ( I personally highly enjoy the beautiful, if potentially inaccurate images of dinosaurs that come from the imaginations of artists and paleontologists, long dead species... fleshed out and in full motion colour. So I am no judge. This beanplating brought to you free of charge by photoshop disasters in conjunction with the guild of lete image analyzer pros....er. first... and 'Shopped.)
posted by infinite intimation at 2:23 PM on January 7, 2010 [7 favorites]


^*note this comment contains no connection in any manner to psd, or the Guild of lete image analyzer pros(GLiaP); Implied, stated or otherwise. And therefore bears no responsibility for the horrifying, and often humorous ridiculousness used by graphic designers of less than full integrity, and possibly less than perfect skill, but surely lacking good sense, and judgment, from across the pop-culture globe.
posted by infinite intimation at 2:31 PM on January 7, 2010


theres a lot to read here -

I think the wider issue is peoples faith in photographys seeming role as modernistic truth telling device - and what happens when that is shown not to be the case.

but then thats just me : )
posted by sgt.serenity at 4:26 PM on January 7, 2010 [1 favorite]


The nature guidelines are interesting because I often cut slices out of gel images and arrange them in a different order so that I can visually compare the things I'm interested in, side by side. Of course when I do this, I leave one or two pixel gutters between them rather than try to make them look like a whole gel again. (Stitching together two gels is a really bad idea.)

Anyhow, in the link with "new" and "old" I really didn't see any of the things they described. The model didn't look lighter, her chest didn't look bigger, her arms didn't look more muscular.

What I see is Edina from Absolutely Fabulous getting all fixated and saying things like, "No one is interested in the sidewalk at the strip mall darling! Oh, I don't know, just blur it out! Oh, and the model. The light is too warm. It looks like morning. No one goes shopping in the morning sweetie. Make her more late afternoony. Well isn't there an afternoony button? It's a state of the art computer, isn't it. (Gets spastic until graphics guy mollifies her by adjusting the model to be cooler.) Yes that's it sweetie, make it bluer still. And that bag. That bag is so last week...."

The difference between Absolutely Fabulous and real life is that in real life Patsy never shows up and distracts the fussy art director with a bottle of vodka.
posted by Kid Charlemagne at 7:53 PM on January 7, 2010


Photographys going to hell in a handbag?

Another key aspect which seems to be overlooked are the basic requirments of digital processing for contemporary forms of output in digital workflows. I think the ethical dust here is that people are still dealing with analogue filmic and print workflows instead of data networked ones.

Processing used to be just called upon in extreme cases when the 'truth' represented by the image needed to be rechanneled or altered. Where as due to the nature of printing and image redistribution on the web or otherwise electronically, now virtually ALL images now undergo processing of some amount before reaching its final audience, its just what the pipleine requires.

Not all 'manipulation' (i prefer to use the term 'processing' for its filmic roots) is done with sinister emotional string pulling of the audience in mind. Most photographers i know are more concerned about getting their images looking right for themselves, as in close to what was in front of the lens combined with their own visions or ideals rather than directly worrying about what their audience will finally see. But then again im in the fine art game as opposed to advertising.

The argument of authenticity of the image in the digital age comes into play when there is some kind of evidence of what is being presented as a package of truth is emotionally divorced from the form/context of how it is being sold. ie in this case 'to bag or not to bag', that is the question.

And finally until this guy fesses up what tools he's using, what delmoi said.
posted by rawfishy at 5:28 AM on January 8, 2010


no photo manipulation thread is complete without at least one dean collins video.
posted by krautland at 8:17 AM on January 8, 2010


« Older Barack Obama as celebrity spokeperson....  |  Lost In The Air: The Jason Rei... Newer »


This thread has been archived and is closed to new comments