Next: Toward Learned Representations Up: Binary Descriptors Previous: ORB
ORB is not the only binary descriptor developed to achieve high computational performance.
BRISK (Binary Robust Invariant Scalable Keypoints) (LCS11) introduces a circular configuration of points sampled around the keypoint. The orientation is estimated using long-distance comparisons, whereas the descriptor is constructed through local comparisons between pairs of points. The result is a rotation-robust and moderately scale-invariant binary descriptor.
FREAK (Fast Retina Keypoint) (AOV12) is instead inspired by the distribution of photoreceptors in the human retina. The points used to construct the descriptor are arranged concentrically, with increasing density toward the center of the analyzed region. This choice produces descriptors that are particularly compact and fast to compare.
AKAZE (ANB13) is derived from the KAZE algorithm and introduces a multiscale representation based on nonlinear diffusion rather than traditional Gaussian filtering. This approach better preserves image structures and contours during the construction of the scale space. The descriptor normally associated with AKAZE is M-LDB (Modified Local Difference Binary).