August 14, 2006
Tim Gasperak pointed me to Photosynth, a provocative research prototype out of Microsoft Live Labs.
The interface strikes me as flickr-meets-David Hockney-meets-Jaron Lanier.
While the research site is short on any real technical details, it does offer a general description of how they develop a "point cloud" imprint for each image that can then be correlated with other images in a collection.
Each photo is processed by computer vision algorithms to extract hundreds of distinctive features, like the corner of a window frame or a doorhandle. Then, photos that share features are linked together in a web. When a feature’s found in multiple images, its 3D position can be calculated. It’s similar to depth perception—what your brain does to perceive the 3D positions of things in your field of view based on their images in both of your eyes. Photosynth’s 3D model is just the cloud of points showing where those features are in space.
This kind of machine-generated metadata seems far more powerful than the hit-or-miss human tagging approach of flickr et al. Now, if they could cross-reference the "point cloud" DNA to GIS data and user-assigned tags, then they might really have something here.
File under: User Experience
_____________________
« Offline
|
Smithsonian Photo Initiative
»
