Scale-Invariant Feature Transform Algorithm

The scale-invariant feature transform, known as SIFT, is an algorithm for finding and describing distinctive keypoints in an image so that the same physical points can be matched reliably across other images of the same scene taken at a different scale, rotation, or viewing angle. It builds a scale space by repeatedly blurring the image with Gaussians at increasing scales, locates keypoints as the local maxima and minima of the difference between neighboring blurred versions, discards candidates that fall on edges or in low-contrast regions as unstable, then assigns each surviving keypoint an orientation from the local gradient directions and encodes the surrounding pixel neighborhood as a 128-number descriptor that stays recognizable even when the image is rotated or rescaled. David Lowe presented the technique in 1999 and published the complete method in 2004 in the International Journal of Computer Vision; the algorithm was covered by a patent that expired in 2020.