Reverse image search as a skeptic tool – with a twist
August 19, 2013 15 Comments
I must confess I’ve been remiss in not blogging about this particular topic earlier. Investigating anomalous photos has always been a skeptic mainstay – for over a century and a half in fact. Ghost photos of one type or another have existed practically since the invention of photography. Those have been subsequently joined by photos of cryptids, UFOs and other alleged anomalous phenomena.
Now that practically everyone has a camera in their pocket all the time (in the form of a mobile phone) photos of this type pop up constantly – along with opportunities to investigate them. And so we need as many skeptics as possible to have some skills in investigating the latest local ghost or UFO photo. There are just too many of them to send them all to Joe Nickell or Richard Wiseman.
Because photos are ubiquitous, and doctoring photos using software is so incredibly easy, tools for detecting photo manipulation (like FourMatch and Tungstene) are becoming more common. But software like that can be quite expensive – out of the range of the average skeptic.
But today I’m writing about something far, far simpler. In fact, it’s an incredibly easy way to detect obvious hoaxes based on stolen or misrepresented photos. It should be in every skeptic’s toolkit, along with a trick for using it that I’ve never seen suggested before.
The Psychic Wayne Case
Although I haven’t blogged specifically about it, I’ve spoken about this technique before. In both my TAM 2012 and SkepTech talks I related the story of Irish skeptic Alan Rice‘s skeptical win over a cheesy psychic program running on late-night TV.
Alan noticed that during “Psychic Wayne’s” program, a seemingly endless series of other psychic readers were promoted via photos and names in the the lower third graphics as Wayne spoke. He reasoned that it was unlikely that the company producing the program actually had that many psychics on call in the wee hours of the morning. So he took screen captures and clipped out just the portion that showed a head shot of one of the other psychics and ran it through a reverse image search engine. He found something interesting.
Most people are familiar with image search – you type a word or phrase and the search engine shows you images that relate to it. You’ve been able to find photos and logos and diagrams this way for years. But a more recent development is reverse image search. Instead of typing a phrase, you upload or provide the URL to an image, and the search engine finds other images which are similar or are exact matches. They can be used to identify people, find the source of a piece of art, and so on – and can even find inexact matches, i.e. altered versions of the original image. Two well-known reverse image search engines are Google Images and TinEye.
In the Psychic Wayne case, Rice was able to crowdsource the task of searching a large number of the other pictures. He and other skeptics were able to use reverse image search to identify that at least 25 of the alleged psychics on call were actually simply stock photos, available for anyone to purchase. Do follow that last link and read Alan’s post about it, as its a great case study of straightforward skeptical activism using online tools.
And best of all, it had an effect. Between the embarrassment of this incident, and a number of complaints to the television station about the content of the program, pressure was brought on the television channel. Before the end of 2012 the show had been cancelled.
The Beast of Trowbridge
Last week a local paper in England reported that a local legendary creature (“The Beast of Trowbridge” – a big cat of some kind) had been photographed. The article displayed the photo and attributed it to an ostensible local couple.
As Hayley Stevens documented on her blog, the photo was quickly found to be a fake. Not only was it not a recent photo (dating from 2007) but it hadn’t even been taken in England. It was a photo of an actual wild cougar (not a cryptid) in Michigan, in the United States.
I was interested in how this hoax was discovered, because Hayley doesn’t specifically say in her post, although she does credit the person who uncovered it. Obviously, reverse image search might have done the trick here. But a quick try using Hayley’s copy of the image doesn’t turn up any useful results. Hmmm.
If you’ve looked carefully at the photos in her blog, you might guess why. The Trowbridge hoax image is left/right reversed from the original from Michigan. It also has a time/date stamp on the lower part of the image (obviously also fake, we now know) that might be interfering with the match. But this does suggest an experiment.
Manipulate Before Matching
Knowing Alan Rice and his cohorts had to crop the head shots to find them on stock photo sites, I figured some manipulation of the Trowbridge photo might help Google match the image. As an experiment, I pulled the original image into a program and left-right reversed it, prior to uploading.
This is easy to do, almost any graphics editing program will have a way to flip or rotate an image. Even if you never edit graphics, you probably already have a program on your computer that will do this. Windows computers come with a program called Paint that can be used, just click the Rotate button on the toolbar, pick an option and save. On a Macintosh the Preview program that comes with OSX can also do this for graphic images – use Tools | Flip from the menu, then re-save the file.
I tried cropping the image to eliminate the time stamp, and also flipping the image. It turns out cropping is not necessary in this case, merely left/right flipping it will return this result from Google Images search:
As you can see, in big bold letters Google states this is a black cougar in Michigan. The upper left of the six visually similar images is the one we’re after – clicking it eventually leads us to this page which is the actual 2007 source of the image. Voila.
Although we worked back from the result, it is clear that a left-right flip is a good choice here, because this photo has an obvious “up” direction. Any other rotation or flip would result in an upside down or sideways image – not likely to match this photo. (Although – you do occasionally see smartphone photos posted in the wrong rotation online).
But it is easy to imagine other cases such as UFOs or ghost photos where there is no obvious up or down, that various rotations or up/down flips might be worth trying. With various combinations of flips and 90 degree rotations, there are eight possible orientations for any given image, as shown in this diagram:
So we might have to create and upload eight files to try all the easy possibilities for a UFO or blobsquatch. (If there is visible text in the image, you could limit yourself to four versions, skipping the ones that would reverse the text). Rotations other than multiples of 90 degrees are possible, of course, but the combinations quickly get out of hand.
We can put all this together to create a best practice recommendation for skeptics.
Skeptics should adopt a best practice of using reverse image search to eliminate the possibility of an obvious hoax for any anomalous photo. My recommendations include:
- Become familiar with both Google Images and TinEye search. (On Google, click the camera icon or drag-and-drop your photo onto the search box to perform a reverse search).
- Both services offer browser plug-ins or bookmarklets to search images with just a click or two – install them. (Google supports Chrome and Firefox, TinEye supports the five major desktop browsers).
- (UPDATE) Chrome users can use this extension instead, which supports multiple search engines. There is also a Firefox version of the addon. (h/t Hayley Stevens)
- Crop the image before uploading to search for only one particular section of the image, as appropriate (Psychic Wayne)
- Images with an obvious “up” direction should be searched as-is and also left-right reversed (Trowbridge)
- Images without an obvious “up” can be searched rotated and flipped, you may wish to try up to 8 versions to get all the possible permutations of simple flips and 90 degree rotations.
These simple steps don’t take much time, and can be done by anyone. They won’t always turn something up, but occasionally they should result in some easy skeptic wins regarding hoaxed or misappropriated photos.
If you are interested in other new techniques, I highly recommend the FourAndSix blog where Dartmouth computer science professor Hany Farid writes fantastic tutorials on digital image analysis techniques.
I also suggest you read Andrew Hansford’s investigation of a UFO photo (just posted yesterday), which he also presented at TAM 2013. It’s a fine example of using online tools and knowledge of digital camera technology to debunk an alleged UFO photo.