A SIFT-Like Feature Detector and Descriptor for Multibeam Sonar Imaging

Figure 4

The SIFT descriptor is created by calculating the gradient magnitude and orientation at each point in the neighborhood of a keypoint. It is weighted by a Gaussian window, indicated by the overlaid circle shown in (a). This normalized neighborhood is divided into sample regions with 8 orientation bins in each, as shown in (b). Therefore, a 128 element feature vector is obtained as the SIFT descriptor of the keypoint.