The Dark Side of Memes


More and more of the information we absorb every day takes the form of images such as this, shared on social media:


Source: Flickr

(That's not a fact, by the way. At the time of writing, no such site as "Memefacts" exists).

There's something both arresting and authoritative about a well-chosen image, with some text over the top, in a nice typeface (that's our corporate typeface, Montserrat Bold, in case you're wondering). Indeed, individuals and groups on both sides of politics have been using such images - let's call them memes, for brevity - to spread their ideas.


Won't someone think of the children?

Children will grow up in this scary, silly, controversial, click-bait world. How can we ensure that they know the difference between what is true and what is false? You could bellow your own political views at them, of course. But that tends to have the opposite effect than was intended on teenagers. What if you could give them the power to spot falsehood themselves? Is that possible?

(Spoiler: the answer is kinda).

Reverse image lookup


Text on top of images often attempts to put the image in a context where it doesn't belong. In the original image I've edited above, the kids are watching the Honda Cog advert, rather than exposing themselves to dank memes. You could find that information by running a reverse image search. There's a few different ways of doing this, which I'll list in decreasing order of usefulness:

  1.  A Google search by image (click on the camera icon, within the search box). Google indexes Flickr, the home for the original image since August 2006, but failed to find the original. Despite this, Google Images is usually my first choice for this kind of task.
  2.  TinEye (the only search which found a match for the original photo, albeit on a different website from where it was originally hosted)
  3.  Bing image search (click on the camera icon to the right of the search button). Bing failed to find the original image.
  4.  Yandex image search (click on the camera icon to the right of the search button). Yandex failed to find the original image.
  5.  ImageRader. ImageRader failed to find the original image, but also took a long time to (fail to) do it.

Google images will allow you to limit the results to a particular time frame. At time of writing, you can do this by clicking on the "Tools" button (to the right of the search field), then from the "Time" drop-down menu which appears, select "Custom range...". Using this, you're more likely to find the original version of the image. To find out if an image has been edited or not, we'll need the largest, highest quality version of it we can find - ideally the unaltered original, straight from the camera (luckily, Flickr lets you download this).

Image forensics

This is where we find out just how misled we've been by Hollywood. In my example image above, I've crudely Photoshopped a Smartphone into the hand of one of my nephews - a Smartphone which didn't exist in 2006. How could we prove that such an edit had been made?

Image Edited is a simple site which allows you to upload an image, and gives you a "yes", "no" or "can't tell" answer. It doesn't actually analyse the image itself (as far as I can tell), but instead, looks to the meta-data of the image file itself.



Meta-data on a file is a bit like those cards you used to get in libraries during the last millennium. It tells you information about the file, but isn't the file itself, exactly. If you take a picture with your phone, for example, embedded in the image is lots of information. This may include when the photo was taken, the brand of phone used, even creepy stuff like the longitude and latitude of where you were standing.

Many social media sites will gladly hoover up this data and even display it on the post about the image. For example, on the page for the original image I used, you can tell that I didn't use flash, the lens was 6.3mm and I used my old Konica Minolta DiMAGE Z1. If you download the original version of the image from Flickr, all of this meta-data is preserved in the file. However, any other size downloaded from Flickr has the meta-data stripped from it, presumably to save on file size. Any of this meta-data can be faked, which means that services such as Image Edited can only offer a definitive guide to the crudest of fake images.

FotoForensics allows you to examine an image for evidence of editing. Obviously, any image with text over the top of it will have been edited to some degree, but it will also show any discrepancies in the image which suggest some element has been removed or replaced. For this to work, the image needs to be as large as possible and the highest quality JPG or PNG you can find. I found it relatively easy to fool FotoForensics when I uploaded a version of the image 1024 pixels wide, but it was only when the image was larger and higher quality that it was clear the Smartphone doesn't belong in that image.

Forensically offers much more fine control over aspects of the image - where as FotoForensics provide only one setting for "Error Level Analysis" (whatever that is), Forensically offers sliders for JPG quality and "error scale" (no idea). Adjusting the JPG quality slider too quickly showed that the smartphone did not belong in my nephew's hands.



Specifically, look at the bright highlight around the smartphone - matched by the text above. This indicates that the phone was not part of the original image.

Likewise, Forensically's noise amplitude ("noise" is the subtle scattering of TV-static-like dots over all digital camera images) also showed that something suspect was happening. This is also something I deliberately tried to fake in Photoshop.



Here, it's the quality of the TV-static like fuzz over the picture which is the give-away. Note that this fuzz is pretty uniform and reddish in colour over flat areas in the foreground. But the screen of the smartphone is duller, darker and has a different "grain".

The hunt for truth

One of the reasons I think that children are so fascinated by the Internet is that it represents so many aspects of their lives - playground, war-zone, safe space, laboratory, library - all in the palm of their hand. Interacting in this environment requires skills which previous generations have been slow to learn, but which are already second-nature to those born in the era of Facebook. And in 2017, we all need to join the search party for the truth.