How does one detect fake images?

[mysts]

Is there software that can scan an image and detect if it has been altered in any way? A pixel here or there, adding clouds to an otherwise
clear sky, changing the color of a car, etc.

[guns1inger]

Not realistically. There might be some sample code or university project prototype code, but nothing for general use. Good work can't be detected that easily, and work so poor that it can is easily spotted by eye.

Simplest method - does the image appear in advertising, magazines, or news services ? If so, it has been manipulated.

[redwudz]

I think it also depends on what you mean by 'altered'. As mentioned, most commercial photographs have been altered, at least color corrected and maybe 'Photoshoped' to improve the image for printing or to remove blemishes or background distractions. If you mean putting a different head on a body then that's usually easier to spot. Look closely at the areas where there seems to be a join. Check shadows and skin tones, etc., for obvious manipulation. But I don't know of any software that can do this.

[lordsmurf]

You'll be able to locate alterations with something like this:
http://www.oceansystems.com/forensic/forensic-Photoshop-Plugins/
It's not a "feature" of the software, but it can be leveraged to reveal alterations from the originals.

Pay site dedicated to topics like this: http://media-geek.com

Professional "CSI" labs/detectives often shoot with Nikon, for this reason:
http://imaging.nikon.com/lineup/software/img_auth/
There's an unconfirmed rumor that this was "broken".

The bigger problem is you can always pay some ******* "expert" to claim anything.
And then juries are being harmed more and more by the "CSI effect".

[mysts]

Thanks smurf

Some interesting stuff here. Since posting my request I found "jpgsnoop" which I will be testing in the next few days.

[handyguy]

Just check the dot density from one area to the next, that is the clue.

[lordsmurf]

Just check the dot density from one area to the next, that is the clue.

No, that's easily fudged if you're good. (Note: I'm good.)

One reason I generally don't assist too much in "how to detect" type topics is because it alerts would-be pests of methods used to catch them. Left to their own devices, many of those folks couldn't Photoshop their way out of a paper bag, and would be none the wiser to their basic mistake (and therefore found by basic detection methods). In other words, finding sloppy work stays easy, it doesn't get harder.

It's a secret in plain site, really. I'm just not overly inclined to announced "hey, look at that!"

[thecoalman]

Depends on the subject, it wont work well for clouds as an example unless you have a very good eye but the first thing to look at is if the shadows are wrong color, missing, wrong angle etc.

Second thing to look for is patterns. If you're doing restorations on images to get the correct color, lighting and consistency you're often going to use the clone brush but if you're not careful you can develop a pattern. There is famous one from Reuters photographer where he was nabbed cloning in smoke...

[olyteddy]

There is also this web app that can possibly tell you where else an image appears: http://www.tineye.com/

145 results for the doctored smoke image above....