12

Can I use Photoshop to detect or find out with 100% certainty if an image is altered or not?

When an image is doctored or manipulated, is there metadata embedded in the altered image so that one can know it is a manipulated image? Or is there some other tool or method that can assist?

I was reading some news about fake UFO photos and this question came to my mind.

6
  • I kind of doubt it, because if I remember correctly the US military has been fooled by photoshop a few times.
    – cutrightjm
    Jun 27, 2012 at 21:43
  • 1
    No way to know for certain. Unless the person doing the editing made a reasonably obvious boo-boo, about all you can do is estimate the likelihood of modification. And even reasonably legit photos are often modified to remove objects in the background, etc. Jun 27, 2012 at 21:44
  • 3
    There may be some metadata left be hind by some photo manipulation programs, but they are not guaranteed to be there or be correct. Any person working to do the forgery well may strip out metadata or put in fake metadata that the file is genuine. After all if they can fake the photo why can't they fake the metadata? Jun 27, 2012 at 21:46
  • 7
    (I always edit my photos to remove the UFOs in the background -- the clutter things up so badly.) Jun 27, 2012 at 21:46
  • 2
    What you want is not possible because you want 100% certainty.
    – Ramhound
    Jun 28, 2012 at 15:02

5 Answers 5

11

Some people (me not included) would claim that Error Level Analysis can prove photo manipulation:

enter image description here

“Error level analysis (ELA) works by intentionally resaving the image at a known error rate, such as 95%, and then computing the difference between the images. If there is virtually no change, then the cell has reached its local minima for error at that quality level. However, if there is a large amount of change, then the pixels are not at their local minima and are effectively original.”
-Neal Krawetz, Ph.D. http://www.hackerfactor.com

3
  • This does not meet the author's 100% certainty requirement..Just Saying :-)
    – Ramhound
    Jun 28, 2012 at 15:04
  • 1
    the service in the first link has been taken down. It looks like hackerFactor.com links to fotoforensics.com though.
    – Griffin
    Nov 24, 2012 at 20:08
  • 2
    Curious; Krawetz’ ELA seems to more perform an edge-detection than differential analysis. @Griffen, errorlevelanalysis.com does still work at this time, and the second link is just part of the quote from Neal Krawetz who wrote a JPEG-quality analysis function which is the basis for his fotoforensics service (and which was used in the ELA site).
    – Synetech
    Nov 25, 2012 at 2:43
10

It depends on what you want to detect. If you want to detect any manipulation at all, then there are some techniques that you can use somewhat effectively. If you want to detect fakes like with bigfoot and Nessie photos, then the options are more limited. While there are actually more signs of a doctored photo, the problem is that they tend to give a lot of false-positives.

When an image is doctored or manipulated, is there metadata embedded in the altered image so that one can know it is a manipulated image?

It's possible, but if someone were going to fake a photo (at least with enough skill for it not be an obvious fake), I would think they would be smart enough to disable or strip any metadata. Even so, the metadata would only indicate that the image had been saved by some program or another, not what was done. For example, someone may open a photo in a program, then save it as another format, compress/optimize it, or resize it. If the program adds metadata to the file, it does not indicate "doctoring", or even necessarily manipulation, only that the program created it (the file, not necessarily the image).

Or is there some other tool or method that can assist?

There are a small handful of programs/plugins that can perform some level of analysis on a photo to look for abnormal artifacts like incorrect edges (as when an item is artificially inserted into a photo), or improper blending, etc. The problem is that most of these can be caused by standard manipulations and so are not reliable ways to detect actual forgeries.

Can I use Photoshop to detect or find out with 100% certainty if an image is altered or not?

You can use Photoshop to manually search for most of the artifacts that the programs/plugins look for. For example, try zooming in to see if you can find image artifacts from the image being re-saved/converted/etc., but that could very well lead to false positives since a non-docotred image would exhibit similar artifacts when saved, compressed, etc.

I was reading some news about fake UFO photos and this question came to my mind.

When it comes to things like this the best and most effective tool is ultimately your own mind. Use common-sense to detect some fakes like photos of Obama eating an alien while riding a unicorn. For more realistic fakes, use your mind to detect things like incorrect shadows or "Photoshop fails":

enter image description here enter image description here

3
  • 1
    You're missing this example :P
    – Bob
    Jun 28, 2012 at 15:36
  • Could you elaborate on what "handful of tools" you're referring to?
    – Griffin
    Nov 24, 2012 at 20:44
  • 1
    @Griffin, most image and quality analysis tools can be used to highlight suspicious artifacts (especially if you have another/original copy). Check out routines like VQMT, SSIM, and PSNR.
    – Synetech
    Nov 25, 2012 at 2:44
6

There is no way of doing this with 100% certainty. If you're being theoretical, people have been manipulating photos since photos started to exist. The pre-war portraits and Civil War photos of Matthew Brady were staged. Shots in the 1840's and 1850s were often composed of multiple shots, since th e film at the time could not handle the brightness level differences - the equivalent of HDR photography. Gustave Le Gray, a photo master in the 1850s, was one of the first to be able to take a sea scene with a single shot. The photo of the Loch Ness monster you mentioned was not manipulated, but was staged. Does that make it more or less real? HDR phots are manipulated, are they more or less real? If you're into the meta thought relationship, check out Errol Morris who talks about photography and realness in depth.

If you're being mostly practical, it's difficult if not impossible to determine this with 100% accuracy. Most photo manipulation software will not leave any metadata as to mark manipulation. There are various ways to run some digital processing on an image, and you may detect manipulations with error level. But if I know the algorithm to detect this, I could fake the data somehow too.

But What is manipulation, what is "normal"? If I boost saturation to match what say, old saturated Slide film would do, is that manipulation or just processing? Making an image B/W is manipulation. Realize that all photography is processing. Even something that seems as simple as a crop can be controversial

Late edit:

Slashdot just posted an article claiming that the World Press Photo Winner was photoshopped, combining multiple files to one image. If you read the linked articles, they explain some of the analysis in depth.

2

Can I use Photoshop to detect or find out with 100% certainty if an image is altered or not?

No, this is impossible. There is nothing intrinsically different about a doctored photo as opposed to an "unmanipulated" photo, so it is always possible to doctor a photo sufficiently well that it is indistinguishable from an original (besides such things as depicting impossible scenes). In fact, the photographing process itself involves manipulation of the image in various ways, and especially digital cameras that we have nowadays extensively alter photos on the fly as they take them.

When amateurs carelessly manipulate photos, they don't go to great lengths to hide their manipulation. There are certain common errors people make that make the manipulation obvious, this is how most "photo verification" works. However, those "rookie forger mistakes" are really not hard to avoid, even for rookie forgers (assuming they have a day or two to kill).

Detecting the forgery is a matter of spotting a mistake made by the forger, which will give it away. But the forger could simply not make any mistakes (or you could miss them), so just because you haven't found any mistakes doesn't mean it's not forged.

When an image is doctored or manipulated, is there metadata embedded in the altered image

Yes, for instance if you take a photo of the sky, then copy/paste a bunch of UFOs in it, Photoshop will add some metadata showing that the image has been altered in Photoshop. However, this is trivial to remove, so just because the EXIF says taken by camera so and so doesn't mean it's genuine.

On the other hand, if it does say it's altered by Photoshop, that doesn't mean anything either. A lot of people have cheap cameras with crappy firmware that can't get color balance right, or takes unnecessarily big photos (for instance many cameras will say they are 12 megapixels, but each pixel is so noisy that you are basically not losing anything by shrinking them to a quarter of the size). A lot of people just routinely open their photos in Photoshop, rebalance the colors, resize (Photoshop has much cleaner algorithms for resizing than many other software) and resave. So it doesn't necessarily mean they doctored it.

A more advanced general strategy is looking at statistical properties of the data. Digital sensors have intrinsic noise (eg. if you take a photo of a flat blue wall, some pixels will have a bit more or less red or green). The pattern of this noise should be uniform throughout a camera: For instance, if you take a sky photo with a very cheap, noisy camera, and then paste the photo of a spinning frisbee from a very high quality, low noise camera, somebody can look at the noise, see the dramatic drop around the "UFO", and conclude it's probably been doctored. Likewise certain characteristics, like sharpness of the image, will also be constant within a photo but vary between photos, especially if taken at different light levels or with different cameras (ie. if you are cutting and pasting from multiple different images).

Again the problem with looking at noise (or light, color, etc.) statistics is that it's very easy for me to artificially add noise that looks "about right", then print the photo from a very high-quality printer, and scan it back with a very cheap scanner (or even photograph it with a camera), then doctor the EXIF and claim it is the original. The editing will be virtually undetectable. This is maybe half a day of work for me, and I wouldn't do it for something like a quick forum thread where nobody will spend more than 5 seconds on the image anyway, but if I wanted to be on the news with a UFO story it's not that much effort.

Error Level Analysis is a similar technique, but has to do with noise introduced when compression like JPEG is applied. Most cameras apply JPEG immediately, and the JPEG adds artifacts to the picture. When you then alter parts of the photo, the artifacts in the altered region will be different. But the problem is, once again it's not very hard to just configure a camera to not use JPEG and take uncompressed pictures, and only compress after you've done all your photoshopping, then ELA will be useless.

Then there are the old-school methods like looking at aberrant lighting, shadows, and so on. Obviously there's nothing stopping a forger from just getting these details right, and with many UFO photos there is so little useful lighting/shadows that you can't do this anyway.

Note that it is also possible to physically construct a fake UFO scene (there have been many notorious examples, like the "pot lid dangling on a string" or what has turned out to be cloud formations) and then take a picture of a it - in which case the photo is of course not doctored in any way (so there is nothing to detect) but that doesn't mean it's a genuine UFO photo.

Or is there some other tool or method that can assist?

Your best bet, if you really wanted to know, is probably to find a professional image analyst who has been doing this for many years, and ask/pay them. Of course you have the problem of figuring out who will do a proper analysis and who will just take your money and lie to you - it's not like there's a perfect litmus test anyway, any analyst will at best give you a probability that the image is altered, and so if you pay someone to the analysis there is little consequence for them if they are wrong.

With forensic image analysis, no two cases are the same, and a lot of it comes down to intuition and skill of the analyst. The skill is something that cannot be gained without extensive experience - learning to do it yourself would probably take you years, and isn't worth your time unless you plan to do it for a living.

I was reading some news about fake UFO photos and this question came to my mind.

With UFO photos in particular, I'd say you are better off just forgetting about the analysis, and simply assuming it's fake. There have been millions of UFO photos, and not a single genuine one. At this point, even if there was genuine documented extraterrestrial activity, it would be impossible to believe photos alone - the well has been poisoned (by frauds and hoaxes) to such an extent that photos are simply useless as "proof".

In US criminal court, the standard of certainty is that guilt must be proven "beyond reasonable doubt". This is a fair standard to go by, and if you apply it to UFOs, proof of photo authenticity beyond reasonable doubt is impossible, just due to the issues I bring up in my answer (there are also many issues I haven't brought up).

In science, the standard is often that there must be at least 95% chance that the null hypothesis (eg. that the photo is genuine) is not true. If you apply this standard, then it is likewise impossible for any genuine UFO photo to meet it outside a few specific circumstances. And even then you can only claim a photo meets the standard in a very naive way - you can at best say "it's very unlikely that this photo has been manipulated after being converted to JPEG for the last time" - which of course doesn't say anything about what happened before the conversion.

These are the two major standards of proof that are commonly used. They are actually quite lax, since both are meant for credible, realistic claims. Genuine photos of UFOs are an extraordinary claim, and should be backed by extraordinarily strong evidence - with a photo alone, this is impossible.

0

a new App - Uproov 'locks' a photo after it has been taken and writes a key into the blockchain, from then on the photo can't be manipulated. Not a bad start to proving if photos are fake or not.

the photo is taken within the app and then immediately encyped and locked. so no time or opportunity to Photoshop it. Isn't that what is needed? If you take it with the phone camera and then use uproov to lock it, you get a different colour ezy-ID (red) which tells you it was an existing photo. Green tells you it was taken 'live' within the App. test it out, better than anything else out there.

2
  • 2
    With that you might prove that you did have a photo with a certain hash at a given point in time, not that it is not manipulated. The latter would be impossible, just imagine taking a photo of a Photoshop window - how would you prove that to be fake given that the screen is of good enough quality?
    – Krumelur
    Jan 11, 2016 at 0:00
  • 2
    This app appears to solve a different problem than the one asked. This establishes no further changes have been made after a given point, not that no manipulation has ever occurred. As such, it's absolutely useless to the original poster. Jan 11, 2016 at 0:34

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .