Picsbuffet (https://picsbuffet.com/#) offers a new method of image search that allows users to easily, visually explore huge sets of images. The images are laid side by side in a “map style” format with similar images close to each other.
All the images of the entire collection are clustered based on general subject, keywords, color and visual similarity. The map can be magnified to 5 levels with a greater or lesser number of similars revealed in each level of search. Search is accomplished by simply moving the mouse across the “map,” and moving the track wheel to enlarge or reduce the view. Keyword search gets the user to an appropriate starting point within the map.
One way to think about this is looking at a map of the United States, then focusing in on Ohio, enlarging to a particular city, neighborhood, street, and finally the house your want. And once you’ve moved as deep as you want to on any track quickly zooming out to look at another location.
Currently the “map” contains one million images sourced from Fotolia. Picsbuffet 0.9 is the first publicly available version and currently is only supported on Chrome and Opera browsers. Future versions will support more images and other browsers as well.
Back in 2012 we reported on
Pixolution, a similar project developed by Dr. Kai Uwe Barthel at HTW Berlin (University of Applied Science). Picsbuffet, which was developed by Barthel, Radek Mackowiak and Nico Hezel is the latest iteration. Barthel says they add more images to their database, once they’ve locked down a server solution powerful enough to efficiently sort large volumes of files.
Using this search technique one thing you become immediately aware of when doing almost any search is the number of “very similar” (almost exact duplicate) images that are available in a 1 million-image collection. I shudder to think what we might get when this search technique is applied to a 50 million-image collection.
To review the pixolution search see
http://www.picslikethat.com/. For further information about picsbuffet please contact Kai Barthel:
onegury@ugj-oreyva.qr