The Glocal Project is a massive contributive artwork. Two months before the launch of the project, we already have upwards of 8,000 submissions from more than 2,000 participants around the world.
One of the most challenging questions has been: how can we make sense of such a large collection of images?
Obviously the first place to start is to catalogue as much information as we can about each image. Some of this information is easy to gather: place, date, place, tags, and other basic information is readily available through Flickr. We’ve also written some simple scripts to record luminosity and to put together a colour pallette for each image. Perhaps most interestingly, we’ve also integrated compositional analysis software, which looks at each image and assigns it a ‘signature’. This signature can then be compared against others in the database to find similar images. This is a very useful tool, since it allows us to find relationships between images that may not have been obvious to human analysis.
I began thinking about these image signatures as a kind of genotype – genetic information that describes each unique image. With that in mind, I wondered wether it would be possible to breed images! The process starts off simply – the image signatures are spliced together at two insertion points:
Sig 1: 1111|111111111111|111111
Sig 2: 2222|222222222222|222222
We then take the child signature and run it through the similarity engine, looking for images in the Glocal pool that matched the child most closely. Happily, this process worked. Below, you can see the three images that result from ‘breeding’ the initial two images. In the offspring, we see the circular element from the parent image on the left in all three images. The most successful child here is the middle one, where we see both the light circular shape from the ‘egg’ and the colour abstraction from the image on the right.
This process can be repeated over generations. In the next image below, I’ve selected the two outside images and asked for images that could be their offspring. In almost all of the child images, we see the consistent circular image in the middle of the frame. There are a few outliers, which may have been imperfect matches – or, more interestingly, which may have picked up on ‘dormant’ portions of the image genotype from previous generations.
We can proceed through these ‘trees’ in a generational fashion, or we can diverge and back-breed. If you take a close look at the image at the top of this post (click to get a larger view), you will see that there is a fair amount of inter-generational mixing.
As this process continues, we can explore the relational landscape that exists in the Glocal pool, and in the process we construct ‘family trees’ which present a possible way in which the images could be related. I imagine an anthropologist, stumbling onto a box containing 8,000 images, might apply similar techniques to make some sense of what stories and histories lay within. These ‘imagined phylogenies’ could be constructed from any database of images, and of course with a larger database the relations would be more clear. Given a large enough database, we could see fairly seamless trees constructed in which the offspring very strongly resemble each of their parents. It may also be possible to apply these techniques to historical databases of images, perhaps providing some useful information about image relationships.
We will be posting the ‘live’ version of this tool very soon. In the meantime, you can see more images in our Glocal Visualizations Flickr set, along with other visualizations that have been produced as part of the Glocal project so far. As always, questions and feedback are welcome and appreciated!