How to tell if its CG or real???

[KLSweetland]

Hi All! -

I work in a different industry, investigating fraud and doing digital forensics. I'm working on a project now and I need to determine whether certain images are CG or real. I have no (repeat... NO) experience doing this. Any suggestions? There has to be some software out there... I figured I would come to the experts and ask.

Thanks so much in advance for any replies. Sorry for the amateur question.

[AJLynn]

You can't really do this in software. Heck, it's getting really, really hard to do it at all. I remember a while back somebody on the forum who was writing a book posted two images of the same building and took a poll on which was a photo and which a render, and it was a very near thing.

 

Here are some images by a member who I think has really nailed down the photorealism recently:

http://www.cgarchitect.com/vb/30918-side-table.html

http://www.cgarchitect.com/vb/31163-sectional-seating.html

http://www.cgarchitect.com/vb/31435-blue-sectional.html

 

For these images, you can tell that if they are photos they were taken by a professional, because they don't have a lot of the problems associated with cheap cameras or inexperienced users - e.g., barrel distortion, "chromatic aberration" and vignette, but if Fran had been trying to defraud somebody I'm sure she would have thought to include these as well, they're not difficult.

 

Probably the most reliable thing you could do would be to hire a consultant who is an expert and have images examined on a case by case basis. If there are going to be court cases, an expert witness is going to be your best bet.

[KLSweetland]

Interesting. I did read up on a court case where the prosecutor basically had an expert that said he could 'just tell' which images were real based on certain elements, and the judge threw him out of court. I have read that there are some statistical programs that can detect the way light is bent by the lense when a photograph is taken, and an image that is computer generated is not subject to the same attributes of course. This is of course relevant to the study of digital forensics because some images are illegal if they are real, some are not if they are digitally created (think: there is no victim). In order for this to hold up in criminal court the threshold is beyond a reasonable doubt, and I don't see how a jury could get there based entirely on an expert's opinion.

[Fran]

A digital photo should have some sort of encoding on it, I think.

[KLSweetland]

Good answer Fran, and some do. Some will blantly state in the 'file header' (first few bits of the digital image) the type of digital camera used to take the photo. Some cameras don't do this though. I'm also concerned that if someone imports the image into photoshop and messes with it, and saves it out again, the file header changes to reflect photoshop rather than the camera. So you can lose track of where the file originated, and therefore whether its real or fake.

[AJLynn]

I have read that there are some statistical programs that can detect the way light is bent by the lense when a photograph is taken, and an image that is computer generated is not subject to the same attributes of course.

 

That's not really true anymore, there is software that simulates these lens effects. Maybe they've got software that keeps up, but if there's some underground image faking community bent on defeating law enforcement's attempts to decide whether their photos are real, they're going to be able to stay a step ahead.

 

Now if it's Photoshop you're thinking of, well, an expert can spot any but the best fake. Everybody seems to be Photoshopping their stuff these days, even the press - I'm remembering recent cases involving Iranian missile photos and one of a soldier who was moved so that he looked like he was giving some people directions when he had actually been pointing a rifle at them.

[AJLynn]

Good answer Fran, and some do. Some will blantly state in the 'file header' (first few bits of the digital image) the type of digital camera used to take the photo. Some cameras don't do this though. I'm also concerned that if someone imports the image into photoshop and messes with it, and saves it out again, the file header changes to reflect photoshop rather than the camera. So you can lose track of where the file originated, and therefore whether its real or fake.

 

This information is very easily faked.

[Tommy L]

wow, what an ace thread. Im assuming you wont be able to post the images? If you do, you will get alot of opinions and some fairly expert discussions going on. If you emial them to me i promise i wont tell anyone......

[BrianKitts]

We make a living out of faking reality, if someone really is trying to create a fraud, there's nothing that can't be digitally replicated, it just depends on their skill level.

 

I would think if you need to catch a fraud you'd be better in photoshop yourself zoomed in and going through the photo with excruciating detail and looking for things like antialiasing breakdown and artifacting, replicated pixel patterns from a stamp tool things like that.

 

But like everyone else is saying, it's getting harder and harder to tell. Makes it worse when you get shows like CSI that do things that aren't possible with photos because the real technical stuff that we would do to dupe a photo just isn't as exciting.

[KLSweetland]

Thanks for all of your replies thus far. Right now I'm on a site at Columbia. I don't have sufficient privileges to post the link, however if you google Trustfoto Apollo and then click demo you can get there. Basically you point to an image and it gives you some statistical analysis of how likely it is that its real or fake.

 

Again, I don't have sufficient privileges to post a link, but if you google It will bring up an article related to what I am working on.

 

Essentially, digital forensic investigators are plagued with a problem related to your field. A few years ago Congress passed a law making it illegal to create and pass around virtual images (that would be illegal if they were real) of children. However the Supreme Court said it was unconstitutional because there were no victims, and threw out the law. So as the law stands, in order to prosecute an individual for illicit images of children you must have identified the child and confirmed that he or she was underage at the time of the photo. As you can imagine, this is ridiculously complicated at times because photos are uploaded/downloaded all over the world again and again, and it can be very hard to trace them back to origin. One of the cases related to this dilemma involved an 'expert' who could only accurately determine 67% of the time whether an image was real or fake. If we could more reliably determine what was real or not, we might not have to actually identify and parade the individual in the photographs thru court in order to get a conviction.

 

I no longer work in law enforcement, but the project I am working on addresses this problem. The better you get at your job, the more difficult mine becomes. And boy are ya'll gettin' good...

 

A few of you asked if you could see the images in question. Obviously the images I'm working on do not depict children illicitly in any way, we are working on random images of all sorts of objects/people/what-have-you (some real, some not) for the purposes of testing. So if I send you an image don't be afraid to open it, the photos are in no way inappropriate.

 

Again, thanks for all the input!

[Lester_Masterson]

I'd say, the easiest way to tell is....

 

look for the Evermotion models in the scene. I mean, my god, they make great models and all, but I see the same damn plants/lamps/books/lemons/bowls/etc. everywhere.

 

2¢

 

(afterthought)

 

Sorry, after reading the article (sorry) I realize the importance of this thread. I initially read the headline and started typing. This shall be a lesson to me.

Edited August 9, 2008 by Lester_Masterson

[litleboy]

I started to think about his few year a go when I saw Final Fantasy. and I thought "what a mess will be when technology is capable to do a virtual clone of let say Angelina and put her on a movie.. no to mention a porno movie"

 

pixels are your friends as long you can get the original pics.

 

we need to be prepared for the future. this is nothing I think.

[Peter M. Gruhn]

Darnit. Two weeks ago I was looking at an article online about just this stuff. See what items I can remember...

 

Light sources, location - stuff from different sources may have different lighting (most likely will unless great care was taken to make them the same). Look for where the highlights fall on curved surfaces, look for what direction the shadows point. Look for how sharp the shadows are.

 

Light sources, kind and count - look in people's eyes in particular. They will reflect the light sources. If this one reflects two soft boxes off to the side and the other one reflects one camera bracket flash....

 

The fella had some software he used which could find cloned bits. Really just a brute force pixel comparison but clevered up so it would only look in the right places. I think.

 

Other things that come to mind:

 

There's the classic repeated stripe. I'll go see if I can find that example at the fail blog. Meanwhile, you might go to the fail blog (http://failblog.org) as a number of the pictures there are faked and there is much discussion. The usefulness of this should be inversely related to the quality of the fakes you run into. ... Can't find the good one.

 

Sometimes with more than one source, one of them will show more JPEG artefacting than the others. "Gee, that guy looks nice and crisp in the middle of that fuzzyblocky scenery."

 

Getting color casts to match can be hard.

 

Cutting one thing out to put in another often leads to fringe around the cutout.

 

Fully CGI scenes mayn't have as rough an edge as they material indicates. Good skin is hard and only recently really even possible. Beginners will let a texture repeat. More and more people are using large full detail photographs for textures these days so it is getting harder to spot bad materials.

[AJLynn]

Oh, you're talking about images of people. This is waaaay outside my area, in more than one direction. What kind of "fake" images are you talking about? Are there people who use legit modes and make them look younger (or claim to do so but it's cheaper to actually use models who are too young)? Making it almost impossible to prove in court that the illegal stuff is actually illegal? Or is it people claiming their photos are 3D renders? (And didn't the court kill USC18.whateveritwas, the record keeping law, which seems to me like a two-edged ruling if I ever saw one, changing the burden of proof?) All in all this is a pretty scary concept.

 

Perhaps one way to do this would be to get an expert to make a digital model of the room in the shot with the lighting and digital standins for the subjects. Then you could pull out information by "render pass" e.g., isolate all the reflections from the lights, the shadows, the specular highlights on skin, etc., and see if they match up. That would identify situations where the context has been faked, or a face has been photoshopped onto a different body, but... is this all done using aging software?

[SandmanNinja]

I think this may be one area that we'll never be able to point a Doo-Hickey at and say for sure.

 

Made Up Scenario:

Two sisters - ages 18 and 14. Dress the younger sister in clothes that are out-of-fashion and she'll look like the older sister in an older photograph.

 

In 3D World magazine, Issue #103, Page 21, there is a full-page ad for a video card.

The character is CGI, but looks pretty darn real.

 

Check out this image here:

 

 

She looks pretty darn real to me.

 

I think as more and more time goes by, as software gets smarter and hardware gets faster, the line will be so blurred that it will be virtually impossible to tell.

 

And I find it troubling that the porn industry is heavily investing in Poser and other 3D technology.

 

http://search.bbc.co.uk/cgi-bin/search/results.pl?tab=av&q=porn%20industry&recipe=all&scope=all&edition=d

[Crazy Homeless Guy]

So what is real, and what is not? ...are advertisements with sexy women real, even though the photos have been rung through Photoshop to remove blemishes, and anything else that is not perfect?

[SandmanNinja]

I think 'real' would have to be defined as 'starting with an actual person'.

[litleboy]

as Einstain said,,,, it's relative.

[3danim8d]

Peter,

 

Are you thinking of the Nova Science Now about Hany Farid

 

http://www.pbs.org/wgbh/nova/sciencenow/0301/03.html

 

 

That was a pretty interesting episode, and could relate the the original posters question.

 

 

 

Jeff

[ronll]

A lot depends on what type of CG you are talking about. If it is a 2D paint image as the woman above, then this is just art and even a good airbrush artist long before computers could make a pretty realistic portrait.

 

If we are talking about a fully rendered 3D image with no 2D pasties of photos, then usually the number of colors registered in a histogram is WAY less than in a photo. The image will compress way more as well because of the smaller palette.

 

If its a hybrid image of 3D with 2D photos pasted into it, then careful analysis of light directions, shadows and reflections can reveal errors.

 

Now as to a photo that has a very minor retouch such as a mole removed then is it really faked? What about a true photo in which a filter was used on the lens? Is that a fake? In fact, what about a black and white photo? Is that a fake because it wasn't really a black and white subject?

[Fran]

I think an image would be considered "fake" from a legal aspect if the intent was to deceive. If a mole was removed in an attempt to convince a jury that a suspect was not who they actually were, then that is a fake.

[ronll]

I agree. But how do you tell intent from analyzing the photo? What if a photographer took a black and white photo of a building because the colors that the architect used on the building were just plain ugly? Was the intent to deceive?

 

Or maybe a less frivolous example yet still subtle, what if a photographer used a filter to make the sky look a little bluer? His/her intent is to deceive you into thinking the sky was bluer than it "really" was. ??

 

I know these seem a little bit silly. But on several occasions I have done expert testimony on view obstruction cases. You would be amazed at the questions the attorneys ask regarding the "truthiness" of the images and photos.

Edited August 11, 2008 by ronll

[Fran]

I agree. But how do you tell intent from analyzing the photo? What if a photographer took a black and white photo of a building because the colors that the architect used on the building were just plain ugly? Was the intent to deceive?

 

Intent is not determined by looking at the photo. The photo, if admitted as evidence in an investigation or court proceeding, would be analyzed to determine its validity. If it was found to have been altered in order to portray a situation as other than it actually is, in order to gain a legal advantage, then the intent was to deceive. This has little to do with aesthetics or artistic choices. Changing a photo from color to black and white does not change the physical shape or location of things. Someone would still be able to identify a person as the same person when shown the same photo in color and in black and white. Cloning John McCain's head on Lindsey Lohan's body and claiming he's a cross-dresser is a different matter. Funny to think about, but the intent would be to defraud. However, if someone did the same thing and merely said, "Hey look! I put John McCain's head on Lindsey Lohan's body! Funny eh?" That would be...artistic expression.

[Crazy Homeless Guy]

I think 'real' would have to be defined as 'starting with an actual person'.

 

This is along the lines of what I was thinking in advertising. It is more effective when the web page is set to mouseover for the before and after, but I don't have time to look up one of those links right now.

 

http://www.nyphotographics.com/retouching/index2.htm

 

or

 

http://m3.torispics.com/piles/?s=baphotoshop#nav-holder

 

.

[Sketchrender]

If the original topic is still up for discussion, wheather fake or not they should be prosecuted.

Intent or not.

Everybody knows it's illegal, so whats the argument here exactly.

Being a parent, i would not let the author away with modeling so called art of somebody so young.

 

I have always thought that there are one or two website that most people know of, that have mostly manga looking models on them, are bordereing on this.

Strange to want to model a young girl.

 

 

Correct me if iI am wrong.

 

If there is not software out there to read tags or what ever it takes to determine if something of this nature is real or not, somebody should write one, i am surprised that the US or other major governments aren't up to grade on this.

 

My 2 cents

 

phil