Making A 2-D Photo 3-D

Some Stanford researchers have been experimenting with using computers to examine 2-D photos and make 3-D models based on them, and you can do it too for free - at least for now.

The approach uses Markov Random Fields (MRFs). Approximating the math for a minute, it's something used in graph theory, which is the study of ... graphs. That is, you've got a lot of points and lines connecting the points. Graph theory gives mathematicians a handle on the properties of these graphs, what you can and cannot do, and how to process things that resemble graphs.

If you think about it, the points and lines can start to look like polygons - in other words, planes. And people have been representing images as complex collections of planes for some time. What the Stanford work does is to look at a photo as planes that relate to each other, and they use MRFs to help process the images and predict what might be deeper in the photo or around a corner, using "supervised learning" to help the program decipher what might be there.

It's a lot of theory, but you can try it out yourself here. The site "takes a two-dimensional image and creates a three-dimensional 'fly around' model, giving the viewers access to the scene's depth and a range of points of view."

Vacation Snapshots Become 3D Models

Researchers at the University of Washington, TU Darmstadt in Germany, and Microsoft Research have an interesting project: they're building 3D models of landmarks out of vacation snapshots. By downloading many pictures from Flickr, they can compare all the different views and have a computer construct 3D surfaces by comparing all the photos, essentially interpolating what surface could offer all the images. There is a lot of sorting, as only a small portion of the photos of any given subject will work for the method. Check the link above for an abstract of the paper and some examples of the imaging, or see this New Scientist article for a more easily digested explanation.