Photos can convey information faster—and sometimes with a bigger emotional punch—than words. But photos are easy to tweak, and in the age of fake news it’s more important than ever to be critical of images presented as fact. This includes feature photos on news stories just like this one (which in this case has been cropped, but is otherwise real).
We’re not particularly good at spotting altered photos. Last week, several news outlets reported results from a study conducted by psychologists at the University of Warwick that found people could only spot fake images 60% of the time. Even when researchers told the subjects the photos were doctored in some way, they weren’t all that good at ID’ing fakery: in those cases, participants could only pick out how they were altered—whether it was a change in the background or shading on the subject—45% of the time.
These results aren’t particularly inspiring for the future of information dissemination, but Hany Farid isn’t discouraged. “The hope here is that people can get good at this with some training,” says the Dartmouth University computer scientist, who is unaffiliated with the Warwick paper.
Farid works in digital forensics, a field focused on identifying the truth about digital records, like visual or audio recordings, that have been doctored. His team specifically focuses on finding ways to identify faked photos, and has developed software that can detect subtle differences in pixels that our own eyes simply can’t. The software is far from being ready for consumer use, but, Farid and his team have figured out some ways to train people to identify when photos have been altered (pdf), or even entirely computer generated (pdf).
Here are some of the tips that Farid thinks everyone could use to improve their photo media literacy:
What’s striking about the University of Warwick paper, Farid says, is that it wasn’t just that people were bad at spotting fake. “It’s that they’re confident in their wrong assessment.” It’s why he thinks that people were only slightly better than chance at recognizing alterations. (You can also take the survey now and rank your own confidence in your responses; the results of the confidence test will be analyzed later.)
Our brains and visual systems are capable of recognizing a lot of detail extremely well, especially when those details help us survive (e.g., distinguishing friendly faces from strangers). But though our brains can quickly process a lot of visual information in ways that would be useful from a biological standpoint, computers are better at spotting patterns we’ve never needed to notice before—like subtle changes in saturation or slightly misaligned shadows. But even just recognizing our visual weaknesses would help make us more critical news consumers.
“You have to take all content that you absorb online with a little bit of grain of salt,” Farid says. Pause to consider the plausibility of what you’re seeing in an image and the story it tells.
Basic training in Photoshop can go a long way.
Images of lone subjects against a background can be easy to fake. All it requires is precisely cutting out the subject from one image and pasting it onto another image. It would be much harder, Farid says, to add in something to a photo that was somehow touching and clearly interacting with multiple other subjects.
It’s also pretty straightforward to manipulate photos to make objects, including people, disappear, just by copying parts of the background and pasting it over the subject in question. (This is called “cloning,” which Farid explained in a NOVA documentary from 2008, around 34 minutes in.)
Blurry or grainy photos—which typically aren’t up to the quality most reliable media outlets use anyway—are also easily manipulated, and should be treated with caution. If images are blurred a little, Farid says, details are obfuscated and our brains may jump to conclusions that aren’t true.
Then there’s the subject matter itself. “Sharks are always fake,” he says. We certainly hope this isn’t true, but it’s understandable that it’s imperative to be more skeptical of images that are objectively harder for a photographer to capture.
Farid and his team have developed software that can detect changes among pixels that our eyes can’t. It can find, for example, repeating patterns pixels that are exact duplicates of each other, which would imply someone had cloned part of the background. It can also calculate exactly where shadows (video) should be falling based on the sunlight using complicated math that can model the 2D image back into its original 3D form.
On our own we’re probably not going to recognize a good fake based on shadow placement alone—but we can certainly spot a bad one. If you see an obvious discrepancy between where the light in the photo is coming from and how the light hits the subject, there’s a good chance that the image has been altered.
Additionally, you can probably detect sloppy editing if you ask yourself what something would ordinarily look like. Something like a treeline in a photo—which we’d expect to be mostly straight in real life—that shows up with a suddenly quick bend could mean that part of an image has been badly placed over another to hide something. Or, a background could appear too perfect: One of the original photos in the survey was edited to show a perfectly full row of trees when in reality, there were some natural gaps between plants.
Sometimes, pictures will distort perspective to give false information. Crops that show 20 or 30 people can lead people to believe that a crowd size was much bigger than it actually was. Consider the photos taken from US President Donald Trump’s inauguration; images from news agencies showed a much smaller crowd than those put out by members of the Trump administration, because the former were taken from farther away than the latter.
It’s likely that you check multiple sources when you read a news story that seems incredible. It’s equally important to look for additional images from a credible source when you are surprised by a photo. If you’re unfamiliar with a news source, Farid advises, it’s good to do a quick search to see where else the story and photo have popped up.
Reverse image searching—using tools like Google Images or TinEye, a free online reverse-image search engine—to find where else an image has appeared can show you if it’s circulating for the first time, or if it’s from a previous event, as Quartz has previously reported.
That said, being more critical of images comes with a trade off: “Increased skepticism is not perfect because it comes with an associated cost: a loss of faith in authentic photos,” the authors of the Warwick study write. If we question everything, we may disregard information that’s important and true.