Oriented FAST and Rotated BRIEF (ORB)

ORB is a fast way for a computer to find distinctive “landmarks” in an image—such as corners on a logo, window edges, or textured parts of an object—and recognize the same landmarks in another image. It is designed for matching images quickly, even when one image has been rotated or viewed at a different size.

How ORB combines two ideas
ORB stands for Oriented FAST and Rotated BRIEF. It combines a keypoint detector with a compact description of each keypoint:

FAST finds candidate corners by testing whether pixels around a central pixel are much brighter or darker. This is far quicker than examining complex gradient patterns.
ORB ranks those candidates with a corner-strength measure, retaining useful and well-spread keypoints.
It estimates each keypoint’s orientation from the brightness distribution around it. A feature on a tilted book cover, for example, gets an estimated direction.
BRIEF describes the patch around the point using many simple brightness comparisons between pairs of pixels. ORB rotates these comparison pairs to align with the estimated orientation, creating rotated BRIEF.

Matching features efficiently
The resulting descriptor is a short string of bits rather than a long list of decimal-valued measurements. Two descriptors are compared with Hamming distance: count how many bit positions disagree. This makes matching extremely fast. ORB also searches an image pyramid, allowing it to find features across several apparent scales. It is less robust than SIFT in severe viewpoint or lighting changes, but its speed and lack of patent restrictions made it a practical classical choice.

Why it matters in real vision systems
ORB supports image stitching, visual tracking, augmented reality, and robot localization. A phone panorama can match overlapping scene details; an autonomous robot can recognize features from earlier frames to estimate motion. In OpenCV, cv::ORB::create() detects keypoints and computes descriptors, while BFMatcher with Hamming distance pairs them across images. Reliable matches can then estimate a homography, align images, or reject incorrect correspondences.

Oriented FAST and Rotated BRIEF (ORB) is a fast, patent-free local feature method that detects image keypoints with FAST and describes them using rotation-aware BRIEF binary descriptors. It supports efficient feature matching across changes in viewpoint and image rotation. ORB matters because it enables real-time tasks such as image stitching, visual tracking, and SLAM with low computational and memory cost.

Imagine trying to recognize a landmark from snapshots taken by different tourists: one photo is tilted, another is darker, and another is taken from farther away. ORB helps a computer find distinctive little visual clues—such as corners, sharp patterns, or unusual details—and recognize the same clues across those pictures.

Its name combines two jobs: Oriented FAST finds those notable spots while accounting for image rotation, and Rotated BRIEF gives each spot a compact “fingerprint” for comparison. This lets software quickly match parts of images, which is useful for panoramic photos, augmented-reality effects, robot navigation, and tracking an object in video.