Glocal Image Breeder

Glocal Image Breeder Interface

Search engines for images tend to be goal oriented – users are typically searching for a specific thing. In this case text-based searches that allow queries like ‘dog wearing a hat‘ or ‘squirrel eating pizza‘ are quite useful. When the goal is to allow a user to browse a base of images without a specific goal in mind, however, this type of system becomes less helpful. How do you present a database of thousands of images for a person to peruse? Obviously a slide-show is out of the question – and presenting page after page of thumbnails is just plain wrong.

With this in mind, I have been developing a series of conceptual search tools for The Glocal Project. These applications have two purposes: they allow the user to wander through the database following non-random paths, and they also act to build relationships between images that can be examined in physical installations of the project in public spaces. We are sharing these tools so that they can be used online, as well as in the workshops that we lead for local highschool students. Eventually, source code all of these applications will be released to the public.

The first of these tools is the Glocal Image Breeder. It allows users to breed images – and look for ‘children’ that may contain common elements from both images. The result is a non-goal-oriented search engine that takes the user through a myriad of possible ‘relational maps’ within the Glocal Database. As additional people use the Image Breeder, more and more relationships between images are exposed. These relationships will then be used by our presentation software to create smarter compositions of images when projected or displayed (you can see a short video of our first prototype installation here). 

This type of relational search tool could be useful in navigating any set of large images. In an art-historical context, it would be interesting to apply this concept to a database of artworks – it maybe that the computer is able to reveal relationships that may not have been previously obvious. Commercially, it could be applied to catalogues of stock-photos, giving customers a new way to find images.


Glocal Image Breeder - Close-up


This tool is still in development – there are a few bugs to iron out and I’m certain the interface could be improved. Any questions or suggestions would be very much welcome. In the meantime, give it a try – the more the engine gets used, the smarter and more robust it will become!

Here are a few hints:

  1. Double-clicking an image will zoom in and reveal a bigger version. Double-click again to return.
  2. Images can be bred with themselves.
  3. If you are having trouble finding a compatible pair in the first set of images, choose the first two. The first 6 images are known pairs, and should always produce good results. 
  4. Use the slider at the top right to zoom in our out.

4 thoughts on “Glocal Image Breeder”

  1. Each image in the database has a ‘signature’ that is assigned to it by the compositional similarity algorithm (libpuzzle). When you choose two images, their signatures are hybridized using a really simple splicing system (no mutation, yet). The engine then looks in the database for images that match the newly-created signature.

  2. this is very interesting! you ever think of using a crowdsourcing tool (like Amazon MTurk) to create the foundation for any mapping DB?

  3. Localguy, that might be possible for future iterations of this concept. For this round, we’re interested at looking specifically at the image-base that has come out of the Glocal Project.

Leave a Reply

Your email address will not be published. Required fields are marked *