SIFT descriptor
A gradient-histogram feature vector built around a keypoint to make matching invariant to scale, rotation, and illumination.
A SIFT (Scale-Invariant Feature Transform) descriptor summarizes the gradient orientations in a patch around a keypoint as a histogram, computed at the keypoint's own detected scale and rotated to its dominant orientation before histogramming -- the scale-and-rotation normalization that makes two SIFT descriptors of the same physical point comparable even when the two images were captured at different zoom levels or camera angles.
Examples
Two SIFT descriptors of the same street sign, one photographed from across the street and one up close, still match because both were normalized to their own detected scale before histogramming.