Feature Point
A locally distinctive point in an image, such as a corner or edge intersection, that can be reliably detected and matched across different views for image alignment and tracking.
A feature point (keypoint) is a locally distinctive location in an image that can be reliably detected across different views. Corners, edge intersections, and texture singularities serve as feature points, forming the foundation for image matching, panorama stitching, 3D reconstruction, and object tracking.
Major detection algorithms:
- Harris corner detection: Evaluates corner-ness via gradient structure matrix eigenvalues. Rotation-invariant but scale-sensitive
- SIFT: Detects DoG extrema in scale space with 128-dimensional descriptors. Scale and rotation invariant
- SURF: Faster SIFT approximation using Hessian determinants and 64-dimensional descriptors
- ORB: Combines FAST detection with BRIEF descriptors. Patent-free and real-time capable
- AKAZE: Nonlinear scale space detection with strong edge preservation
Processing involves detection (locating keypoint coordinates) and description (encoding surrounding patterns as vectors). Matching descriptors between images identifies correspondences. RANSAC removes outliers for geometric verification. Deep learning methods like SuperPoint and LoFTR now surpass traditional approaches in challenging conditions.