[Object Recognition] SIFT Algorithm: Theory
SIFT (Scale-Invariant Feature Transform) is an algorithm for detecting and describing local image features. The algorithm searches for extrema across scale space and extracts position, scale, and rotation invariants. These keypoints remain stable under changes in illumination, affine transformation, and noise. The main steps are (1) scale-space extrema detection; (2) keypoint localization; (3) orientation assignment; and (4) keypoint description and matching.