Image Doctoring Is Tough To Spot, Even When We're Looking For It

An anonymous reader writes with a link to The Stack's look at study that examines the human capacity to detect (or to overlook) manipulation in images. About 400 volunteers looked at images which had been digitally altered by erasing elements, by replicating parts the image, or by pasting in elements from other images. Less than 58 percent of the alterations were detected, even though the volunteers knew that's what they were to look for. The article says "While its conclusion – that we are not very good at identifying doctored photos – is predictable, it's the type of 'fakes' that deceive us which are most interesting." Spoiler: Erasure is much harder to spot than image splicing.

Comments

My worry is the credibility loss of visual records

[He Who Has No Name]

A couple decades from now, we're going to have generations that have never known an era when it was technically or logistically difficult to convincingly revise AND distribute photos and videos.

Beyond simply telling that stuff has been tampered with or invented wholesale, I'm really worried this is going to lead to a loss of credibility and gravitas of photos and videos of historic events.

It's going to get ugly when generations start denying and rewriting history because they lost trust and belief in the credibility of the medium used to preserve its records.

False premise

[DNS-and-BIND]

I know these sorts of articles are wildly popular these days, showing HOW DUMB ALL OF YOU OTHER PEOPLE ARE. I understand, it's very reassuring to see yourself placed in the top position where you can shit on everyone else. Thousands of years of tyrannical human elites agree with you. But you don't need everyone to see it. It just takes a single person to spot that something is wrong, point it out, and the viral internet takes over from there. That's how Tom Brokaw's fraud was exposed, someone said, "Hmm, that looks just like MS-Word" and then made the animated .gif that changed the world. Thinking that everyone needs to be a Photoshop expert is just naive and misanthropic. Reuters was also caught red-handed altering photos to conform to their narrative. It just takes one person to utter the sacred phrase "Hmm, that's funny".

Easy way to make tidy shops.

[Anonymous Coward]

One easy way I found to make some tidy shops is resize the image to quadruple its resolution and edit it there.
Then, when you are done, resize it back.
It smooths things together pretty nicely on a sub-pixel level that is usually undetectable even to machine-scanning techniques.
Of course, if you have the original image, it will be trivial to spot the edit regardless, but it will be harder to figure out which one is the original image.

Another I have been experimenting with is adding in grainy-looking texture from awful quality CCDs at that quad-res layer.
It is far easier to do this at the quadruple resolution because the botched noisy pixels will be enlarged quite considerably.
Now instead of nearly breaking your arm trying to be overly precise, you can easily just draw a circle and apply whatever color or filter to it. (in another layer, of course!)

Sooner or later, with smarter editing software, it is going to be damn near impossible to tell an image is edited without the original.
People in any position of power will be able to, quite literally, rewrite history. (not that this has changed anything, this has been true for thousands of years)

Re:Easy way to make tidy shops.

[Anonymous Coward]

Yeah, this is pretty good-sounding stuff.

To some extent I've used that "scaling up" trick myself, though that's usually if I want to tidy up a soft-edged logo or whatever. Simply increasing the contrast at low resolution gives jaggies. Scaling up massively, *then* performing various filters before increasing the contrast to give a very sharp edge looks much better when it's scaled back down.

I often add noise to edited portions of an image that have no grain or noise, so that they more closely match the appearance of the original image. The noise in Photoshop is too obviously fake on its own, so I usually add NR to it afterwards (you have to increase the initial noise so that the NRed version gives you the level you want). Sometimes rescaling the noise layer is required.

I do this so that the edits don't look bad, not so much to fool people (a lot of this is for (e.g.) Wikimedia Commons where I like to be explicit about any retouching done anyway).

Photo's subject matter is important for humans

[sleepypsycho]

Look at the high variance images at the top of page 9 of the original article. I finding unsurprising at all the people can't tell if a sand dune was smoother or if erasures occurred on image of pine tree branches against pale blue sky.

One documentary on the development of new currency said portrait sizes were increased because of our ability to notices small variation in faces. I don't have the reference for this, but my personal observations match the claim.

My wife and I are both artists with masters of fine art in painting. For years, bad effects in movies will jump out at us. We will sometime refer to the "cgi cast of thousand" in egregiously offending movies. For some years now my ability to identify cgi inanimate objects has almost disappeared. In modern movies I almost never have a cgi object jump out at me. I notice that the cgi animation of people is similar improving, although glaring problems still appear. I am sure it has both to do with complexity of the physics problems and my own visual capability and nature focus of attention. I except it is few short years until I almost never notice any cgi modifications.

(Most) Humans Are Easily Fooled by Digital Images

[ThaumaTechnician]

From my experience, professional photographers, photo lab technicians, and their ilk, who spend lots of time looking photos and their technical aspects (sharpness, lighting, colour, pixelation, etc, etc, etc), would easily spot the, to me, obviously edited photos. Case in point: the top photo in the The Stack article - clinton-fake-photo-832x333.jpg - is so obviously faked that I'm surprised that anyone would be fooled by it.

First dead giveaway: Clinton's head is illuminated from the right-read as is Mandela's(?) head. De Angelis's head? From the top-front. (I can't believe that De Angelis's victims didn't spot that fakery.)
Second less-dead giveaway: the pixelation isn't quite the same. The imposter's pixels are larger, not by much, than Clinton's - at least in the sample from the article.

Re:That's not good thing for us ?

[AK Marc]

The human mind demands "why". When there is no "why" the mind has greater trouble accepting the event ever happened. So for deleting an inconsequential flier on a board, or duplicating a cistern in a photo of cisterns, there is no "why". But inserting a bus into a busy street, someone could think of a "why". To hide something behind it, or to make the streets look more busy (as done with many movies, including Fury Road).

When the only "why" is "to see if you could spot it" then the why has no real meaning, so the fake will be harder to spot.

I think they stumbled onto something that tells the inner workings of the brain, but not the workings they were looking for. Repeat the test telling someone "This photo was altered to [something]" as the hint, and I think they recognition rate will increase. And I think that if the "why" is pre-shared, an fMRI would show a greater portion of the brain activated for the detection.

Now if only I was independently wealthy, I could fund these more useful studies.