Typically keypoint detectors are designed to capture features with high repeatability if exposed to noise or to changes such as illumination, view point, or deformations. These detectors varies depending on the method of extraction. Some of these methods are edge-based, corner-based, or blob-based A.
are robust to geometrical and non-geometrical changes. Normally, the detectors are evaluated separately by calculating the repeatability score along with changing scale, rotation, viewpoint, illumination, or compression ratio of a of a reference image. Descriptors are associated with the detectors, usually they are evaluated and sometimes with the associated descriptors. The matching score, precision, recall, F-measure, usually computed for the both; the detector and its descriptor. Can evaluate descriptors fixed detectors?. Given two images, usually, the keypoints detected from the reference image are compared to the keypoints detected from the query image by comparing their descriptors. Most likely, the descriptors pose important features and are computed from edges, corners, or regions related to the detected keypoints. Therefore, keypoints are referred to local features in literatures.
The descriptors are compared using some distance or similarity measure for finding the corresponding keypoints or regions between images.