SIFT is an algorithm in computer vision to detect and describe local features in images (“interest points”). The goal is to identify known objects in new images. Applications include robotics, navigation, image search, 3D scene modeling, a lot of what we’ve been talking about in Representing Earth…
The algorithm is rather technical, and more detailed descriptions are readily available for those interested, but the gist of it is to find repeatable and invariant features that can be matched between images.
- Each SIFT feature is described by its location, orientation, scale, and a keypoint descriptor (accounts for shape distortion and changes in illumination)
- New images are recognized by computing SIFT features on the input image and comparing these to a database of SIFT features from known (“training”) images.
Note: Amazon recently bought a company called SnapTell, which had implemented a recognition system using their own features that presumably were modeled on SIFT, but didn’t use SIFT code. Google Street View also uses their own image matching, probably similar to SIFT, but using their own code.
4D Cities project