Search Images by Navigating Through a Network
Congratulations to our CVO Prof. Dr. Kai Barthel and his Visual Computing research team at Hochschule für Technik und Wirtschaft Berlin. They were awarded last week at ACM Multimedia Conference in Nice, France for their demonstrator of a new graph-based image search system.
It’s an impressing project and a great inspiration for our team at pixolution since we’re always excited about new image search concepts. That’s why we would like to share this idea with you.
Wikimedia Commons Images
The demonstrator is called Wikiview and contains over 12 million freely usable images from the Wikimedia Commons website. Wikimedia itself only offers a simple keyword search, where many images are poorly tagged or not tagged at all, which makes a targeted search very difficult. With Wikiview these images can be searched by keywords, colors, uploaded sample images or — and this is the cool thing — navigating.
Search by Navigating
The new graph-based image search concept deals with a major problem of common image search systems: As the number of images increases, image search systems typically become slower and slower. With Wikiview each image is linked to similar ones which eliminates the time-consuming search for similar images. Instead the user traverses the network of linked images which are then visually sorted and displayed.
Representing images in a network means that the relationship between images is baked into the data structure. Now, searching is navigating the network.
What may sound complicated is very easy for the user: Search results are presented nicely sorted on a sphere and by dragging or zooming it with the mouse similar image concepts can be explored visually. This brief tutorial shows the use of Wikiview.
We think Wikiview is a great way to visually explore huge image sets. If you haven’t tried it yet, be sure to try it yourself. Happy navigating:)
Originally published on pixolution.org/blog.
