Microsoft previews Photosynth|
(Tuesday, August 1, 2006 - 14:05 EDT)
Microsoft Live Labs has released a preview of its upcoming Photosynth application, apparently created in conjuction with the University of Washington.
Photosynth defies a description that really does the technology justice, but in essence it allows you to takes large collections of related photos, automatically find similarities between them, and then display them in a three dimensional space that is reconstructed based on the image analysis.
A video on Microsoft's Live Labs site demonstrates the process nicely, while a second video (choose the long version) on the University of Washington Computer Science and Engineering site takes a little deeper look at just what the technology can do. Perhaps most impressive is the live Java demo of the technology on the UofW website. It doesn't include all of the features seen in the videos, but the ability to control everything yourself and see the results will give you a nice feel for how it all works.
Summarising the technology in a little more detail, Photosynth will search each image for recognizable points that are repeated across multiple images in a group. It then uses the locations of these points to assemble a "3D point cloud" of the entire location / object, along with the estimated locations of each camera as the photos were taken. The viewer is then given an overview of the scene, showing both the data points and camera locations. Clicking on a camera location takes you to the image captured by that camera; selecting an area in the point cloud takes you to the camera view that best covers that area. From there, you can jump to other camera locations, or select further areas in the point cloud. You can also step sideways to adjacent camera locations, or backwards to a location behind you.
The UofW video shows further abilities including the possibility of tagging areas of images with a description, and automatically have those same areas tagged in other images where they're clearly visible. It also suggests that it will be possible to import tags from communal photo sharing sites like Flickr, and shows off other ways of rendering both the point cloud, and the transitions when moving from one image to the next.
This is all truly impressive stuff, and we're already excited to see more on the technology. Readers take note - before watching these videos you may wish to place a padded object in front of you. If your reaction is similar to ours, you may otherwise find yourself bruised as your jaw meets your desk at high speed. ;)