Show Hide image

You shouldn’t believe your eyes: how to identify fake images online

In today’s post-truth society, it’s more important than ever to verify the images you share online.

A photo of Jeremy Corbyn walking through a train carriage of empty seats is many things, but it is not faked. The image, taken from CCTV footage of the 11am train from London to Newcastle on 11 August, hit the headlines last week after Virgin refuted the Labour leader’s claims that he couldn’t find a seat on its service.

Due to the highly-politicised nature of this image (yes, really) many people were quick to question its legitimacy. “It's in the must be true! Cos technology can't enable anyone to airbrush or photo shop things like seat tickets yet,” wrote a commenter on the New Statesman’s own Facebook page, with another noting that Virgin took two weeks to release the images.

It is true that in today’s post-truth society we should question photographs that we see online, particularly those that arguably have an agenda. Many people find it easy to Photoshop an image, but you don’t have to create a fake image to be complicit in the damage they cause. After Hurricane Sandy in 2012, researchers found that 86 per cent of tweets spreading fake images were retweets, not original tweets. Before you share an image online, it is important to analyse its legitimacy. Great. But how?

“Let me emphasise that there is no single test that will tell you everything you’d want to know about the provenance of an image,” says Kevin Connor, the ex-vice president of product management at Adobe and founder of Fourandsix Technologies, a company dedicated to analysing the authenticity of digital images.

“Instead, you really need to play detective, searching for a variety of clues that can provide you with an indication of where an image came from and what has happened to it. One of the first and most useful things you can do is to perform a reverse image search, using a service like Google Images or TinEye. You may want to use both, because they sometimes provide different results.

“If you find any matches, look for the earliest appearance. If the earlier version looks different than the version you have, then your version is probably modified. Even if the image doesn't appear modified, you may discover that the image is not what it claims to be. For example, it's not uncommon for an image to appear on social media claiming to be of a crowd in a recent protest, but reverse image searches then reveal that the image was actually taken in a completely different city years earlier.”

The results of a Google reverse image search on the alledged Hurricane Sandy photograph

Connor became interested in analysing the authenticity of images because he spent over 15 years driving the Photoshop product line at Adobe and found that journalists and law enforcement professionals frequently questioned him on how to tell when images were manipulated.

“I didn't have great answers for them,” he says, “but then I became aware of the research of Hany Farid from Dartmouth College. He was the first person to seriously study algorithms for authenticating photographs. When I was leaving Adobe and deciding what to do next, I decided to team up with Hany and see if we could productise his work.”

One of the main products resulting from this collaboration is, a website that allows photographers to certify that their images are original, unmodified files from their camera. This is useful if you want to prove to an online beau that you aren’t catfishing them, or that the trainers you’re selling on eBay are the real deal. Unfortunately, the service is limited.

“For the typical files you find online or in social media, izitru won't be very useful, because these files are generally not the original files captured from the camera,” says Connor. “For example, Facebook and Twitter never serve up the original image file that someone uploads. It always gets re-saved at a smaller size, with most of the metadata stripped out. Izitru is most useful if you know the person who is sharing the photo, and you can ask for the original. They can either upload it to izitru themselves as proof, or they can send it to you so that you can upload it.”

Reverse image and izitru are both accessible and reliable because you don’t need to be an expert to use them. When it comes to image forensics and analysing the make-up of an image, however, Farid warns me: “There is little the average person can do to reliably determine that a photo is fake.”

“If you do know more about image editing software, then you could start looking for clues within the image – traces that the image editing tools leave behind, or mistakes that the editor made,” says Connor. “This is difficult, though, and many people reach inappropriate conclusions.” This was best illustrated during the 2013 World Press Photo of the Year, when Dr Neal Krawetz, a photography forensics expert, argued the winning image was a composite of various photos. Connor, Fadid and Eduard de Kam from the Nederlands Instituut voor Digitale Fotografie were called in to forensically analyse the photo, and verified that it was real.

De Kam says the easiest way to analyse an image is to get your hands on the raw file, something that can often be difficult. “If there is no raw file, the highest resolution is what I have to work with,” he says. “That means slowly moving through the image trying to find errors, a slow and not very certain process.

That’s not something you or I can easily do, and the meme “This looks shopped, I can tell from some of the pixels” mocks those who think they can. There are, however, a variety of services online that claim to be able to automatically detect fake images. Connor warns against many of these.

“In collaboration with Hany while I was still at Adobe we demonstrated a plug-in that would automatically point out regions in an image that might have been cloned, he says. “We chose not to release this, however, because we felt that in novice hands it might lead people to incorrect conclusions. That's because most of these techniques are not 100 per cent conclusive, and you still need to make an educated judgment about the results.

One of the most common ways websites and internet sleuths will claim to delegitimise an image is through error-level analysis (ELA). ELA shows the differing levels of compression throughout an image, in a way that ostensibly highlights where an image has been changed. For example, if I run this still of Harry Potter through, it will rightly highlight the differing compression levels on the image of Danny DeVito I have pasted onto the Dursleys’ heads in their family portrait.

Yet although this seems pretty irrefutable, Connor warns against ELA. “We simply don't place much value in that test because its results are especially inconclusive,” he says. “People like to use it because it's available online for free and it provides visually interesting results, but these results don't have a lot of value.

“Firstly, it just provides you with a pretty picture without giving you much indication of how to interpret it. Yes, you're told to look for areas of the image that look different than the rest, but how is someone supposed to judge what is ‘different’?

“Additionally, because the technique is testing for something that's only tangentially related to image editing, it's prone to producing unacceptable levels of both false positives and false negatives. In other words, it's not very difficult to find edited images that don't product suspicious results in ELA, and it's also not too difficult to find unedited images that do produce suspicious results.

“At best, ELA might be useful for directing your attention to certain areas of the image that may deserve future scrutiny, but you shouldn't make any final conclusions based on ELA alone.”

When it comes to analysing an image then, it might be best to use the cheapest and most user-friendly computer at your disposal: your brain.

“It is harder to fake an image then to write total nonsense,” says de Kam. “So we should always keep thinking whatever we see, whatever we read, how much sense does this make?”

Many fake images that spread online aren’t fake in the sense that they’ve been manipulated, but fake in that the stories attributed to them are false. The first step, therefore, is to apply critical thinking. Are there any other images that can corroborate the one you’re looking at? For example, by finding other images from the day Hillary Clinton was photographed next to a woman wearing an "I'm with stupid" t-shirt, we can easily see the image was faked. 

Another important question to ask is whether the source reliable, and what agenda they might have. Once you've exhausted your brain, there are a multitude of low-tech services that can help you find answers, such as and Twitter accounts like @HoaxOfFame or @PicPedant, which are dedicated to uncovering fake images and stories.

Being sceptical and asking questions are therefore most people’s best options. The Defense Advanced Research Projects Agency (DARPA) has recently begun a large research effort into image forensics, and, with Hany Farid working on the team, might develop a way to automatically detect image manipulation.

“For now, though,” says Connor, “it's still some tricky detective work.” 

Amelia Tait is a technology and digital culture writer at the New Statesman.

John Moore
Show Hide image

The man who created the fake Tube sign explains why he did it

"We need to consider the fact that fake news isn't always fake news at the source," says John Moore.

"I wrote that at 8 o'clock on the evening and before midday the next day it had been read out in the Houses of Parliament."

John Moore, a 44-year-old doctor from Windsor, is describing the whirlwind process by which his social media response to Wednesday's Westminster attack became national news.

Moore used a Tube-sign generator on the evening after the attack to create a sign on a TfL Service Announcement board that read: "All terrorists are politely reminded that THIS IS LONDON and whatever you do to us we will drink tea and jolly well carry on thank you." Within three hours, it had just fifty shares. By the morning, it had accumulated 200. Yet by the afternoon, over 30,000 people had shared Moore's post, which was then read aloud on BBC Radio 4 and called a "wonderful tribute" by prime minister Theresa May, who at the time believed it was a genuine Underground sign. 

"I think you have to be very mindful of how powerful the internet is," says Moore, whose viral post was quickly debunked by social media users and then national newspapers such as the Guardian and the Sun. On Thursday, the online world split into two camps: those spreading the word that the sign was "fake news" and urging people not to share it, and those who said that it didn't matter that it was fake - the sentiment was what was important. 

Moore agrees with the latter camp. "I never claimed it was a real tube sign, I never claimed that at all," he says. "In my opinion the only fake news about that sign is that it has been reported as fake news. It was literally just how I was feeling at the time."

Moore was motivated to create and post the sign when he was struck by the "very British response" to the Westminster attack. "There was no sort of knee-jerk Islamaphobia, there was no dramatisation, it was all pretty much, I thought, very calm reporting," he says. "So my initial thought at the time was just a bit of pride in how London had reacted really." Though he saw other, real Tube signs online, he wanted to create his own in order to create a tribute that specifically epitomised the "very London" response. 

Yet though Moore insists he never claimed the sign was real, his caption on the image - which now has 100,800 shares - is arguably misleading. "Quintessentially British..." Moore wrote on his Facebook post, and agrees now that this was ambiguous. "It was meant to relate to the reaction that I saw in London in that day which I just thought was very calm and measured. What the sign was trying to do was capture the spirit I'd seen, so that's what I was actually talking about."

Not only did Moore not mean to mislead, he is actually shocked that anyone thought the sign was real. 

"I'm reasonably digitally savvy and I was extremely shocked that anyone thought it was real," he says, explaining that he thought everyone would be able to spot a fake after a "You ain't no muslim bruv" sign went viral after the Leytonstone Tube attack in 2015. "I thought this is an internet meme that people know isn't true and it's fine to do because this is a digital thing in a digital world."

Yet despite his intentions, Moore's sign has become the centre of debate about whether "nice" fake news is as problematic as that which was notoriously spread during the 2016 United States Presidential elections. Though Moore can understand this perspective, he ultimately feels as though the sentiment behind the sign makes it acceptable. 

"I use the word fake in inverted commas because I think fake implies the intention to deceive and there wasn't [any]... I think if the sentiment is ok then I think it is ok. I think if you were trying to be divisive and you were trying to stir up controversy or influence people's behaviour then perhaps I wouldn't have chosen that forum but I think when you're only expressing your own emotion, I think it's ok.

"The fact that it became so-called fake news was down to other people's interpretation and not down to the actual intention... So in many interesting ways you can see that fake news doesn't even have to originate from the source of the news."

Though Moore was initially "extremely shocked" at the reponse to his post, he says that on reflection he is "pretty proud". 

"I'm glad that other people, even the powers that be, found it an appropriate phrase to use," he says. "I also think social media is often denigrated as a source of evil and bad things in the world, but on occasion I think it can be used for very positive things. I think the vast majority of people who shared my post and liked my post have actually found the phrase and the sentiment useful to them, so I think we have to give social media a fair judgement at times and respect the fact it can be a source for good."

Amelia Tait is a technology and digital culture writer at the New Statesman.