One of my original aims when I started collecting selfies was to build up a taxonomy of sorts. It interest me that, like most genres, each photograph is unique but it will fall into a grouping. People will adopt certain poses, sometimes stylistic, sometimes dictated by needing to be both photographer and subject, and those poses repeat again and again.

I decided to look for something like Google Image Search that I could run on my Mac. I wanted to take a selfie and have the computer find similar selfies. I haven’t yet found the perfect app to do this as it seems the reason Google Image Search works so well is because Google powers it with more than a Mac Mini, but I did find Unik, an app made for finding and deleting duplicate images on your hard drive but which uses some visual recognition to find images that have been edited or resized in some way. It recommends you set the similarity to 98% but allows you to reduce it to 80%.

I decided to feed it all 50,000 of the #selfie tagged images I’d collected, including those I hadn’t sorted through yet, to see what it could find.

At 98% all I got were actual duplicates, which was useful from a housekeeping perspective but not very illuminating. As I moved down the scale groups of allegedly similar photos emerged, mostly those taken in the dark so the colours were similar. Again, not very interesting. As I got to 85% the groups lightened up and had more variety and I started to see that this definitely wasn’t a face recognition algorithm. It was looking for (at 85% quite fuzzy) shapes and was quite ignorant of what those shapes were.

But the patterns were definitely emerging. Here’s a random screengrab of the results.

With the computer realising its limitations it was time for the human to take over. I scanned through the groups and started picking out pairs of selfies that, to me, actually looked similar. I wanted the same pose, the same position of the face. I grabbed these pairs and made them into animated GIFs, and because they were animated GIFs I created a Tumblr to house them.

You can see all the pairs I found this evening at doppelselfies.tumblr.com. This one is probably my favourite.