Identifying Manipulated Images

Jamie found a cool story at MIT Tech Review. (As an aside, it sits behind an interstitial ad AND on 2 pages: normally I reject websites that do that, but it's a slow news day, so I'm letting it through.) Essentially, software is used to analyze light patterns in still photographs. Once you can figure out where the light sources are, it becomes a lot easier to determine if an image has been photoshopped.

Detector == Quality Control

[Anonymous Coward]

People who manipulate images will use these tools for quality control: When the fabrication passes all tests, it is ready to be released.

Colombo did this on his 1970's TV show

[jimwelch]

Someone wore a photo mask and tripped a speed camera to give their partner proof that they were across town (LA) at the time of the murder. He noticed the shadow under the nose was wrong by comparing previous and following pictures from the same camera.
I am not sure which episode it was. Peter Falk as Det. Lt. Colombo

weak

[gnudutch]

this method is way better

Forensic Analysis Reveals Al-Qaeda's Image Doctoring [slashdot.org]

Finding Photoshopped Pics for Fun

[Dan East]

Does anyone else have a habit at looking at pictures and trying to see how they've been manipulated? These types of pictures are rampant in advertising. Pick up any magazine and start looking, and the poorly edited pictures will jump out quickly. The more professionally edited pictures have much more subtle problems, and can take a bit of poring over to find. Many product images (on packaging and in catalogs) are the same way, and are usually the worst edited of the bunch. Some things I look for:

• An object rubber-stamped in multiple places. Each copy is identical, which gives it away. They are often scaled, rotated or mirrored to make them look more unique.
  
• Lighting and shadows, which is what the algorithm in this story deals with specifically.
  
• Focus. Often multiple objects will be in focus at varying distances impossible with a single shot.
  
• The same image of a person is used in multiple shots. This is most prevalent in product images in catalogs.
  
• Poor masking, where edges of objects are over or under processed, either clipping part of the object (hair can be particularly tough to do), or showing some color edges from the original background.
  

Anyway, that's just the geek in me I guess, because I really do enjoy finding flaws in images. What I hate is an image that has a sort of surreal perfection to it that I know must be composited, but I can't find any smoking gun.

Re:Adds a step for the photoshoppers

[CambodiaSam]

Light sources have always been a pain for me when photoshopping. I'm not surprised that it's the key to this software, as it tends to be the most difficult aspect of manipulating an image (I'm a slightly more than casual user, but not a graphic designer). Light completely changes the color structure and I end up spending an inordinate amount of time trying to redo hues and fix shadows that don't line up.

If there's a plugin for helping me with that part of the struggle, I hereby scream to my fellow slashdotters to please fill me in!

speaking of which

[circletimessquare]

did anyone else catch the blog in the new york times about the fenton photographs [wikipedia.org]?

apparently this guy took some photos of some cannonballs in the crimean war that became famous as a poetic commentary on war. this documentary filmmaker, errol morris [wikipedia.org], has gone completely unhinged obsessive compulsive over whether or not the photos are fake and/ or manipulated. it's utterly fascinating, and a little weird, to see so much time and effort devoted to these photos. specifically, cannons and shadows. utterly esoteric and thorough. he also expands into the larger topic of the history of manipulated politically sensitive photos. makes for a good read, especially if you are interested in pre-photoshop image manipulation

check it out, talk about thorough [nytimes.com]

way better?

[CrazeeCracker]

The interesting thing about that is... The code used in the article the OP linked to features the following lines:

Revision history: This code has been stripped out of imgana by Hacker Factor Solutions. (Imgana does much more than quality analysis, but that's all that is being released right now.)

Said program by Hacker Factor is also mentioned in TFA as a more basic approach to checking whether or not an image has been manipulated. I'll leave you to judge what this means.

As an unrelated sidenote, Hacker Factor features a very interesting javascript that guesses the gender [hackerfactor.com] of the author of a block of text (>300 words). Thus far, I've found it to be eerily accurate.

Re:Detector == Quality Control

[jd]

Detecting discrepencies between light sources is something that can be written as an algorithm and should increase in complexity linearly with the product of the number of light sources and number of pixels. Determining the correct value for pixels is something that must be written as a herustic and will increase in complexity exponentially with the number of objects (including light sources and non-visible objects) in the system. A herustic is not guaranteed to finish. Ever. Even if it always did, because the problem is exponential, far more images can be checked than can be fixed up.