keypoint descriptors

Compact numerical representations that encode the local appearance around detected keypoints, designed to be distinctive and robust to geometric and photometric transformations. These descriptors act as "fingerprints" for visual features, enabling reliable matching across viewpoint changes, scale variations, rotation, and illumination differences. Common types include gradient-based (SIFT, SURF), binary (ORB, BRIEF), and learned descriptors (SuperPoint, D2-Net). Key properties: discriminative power, robustness to transformations, computational efficiency, and dimensionality. Applications: feature matching, object recognition, visual SLAM, image stitching, and structure-from-motion.