Detection and Localization of Point Landmarks

Abstract

Geometric image features denoted as landmarks are preferred features for a variety of computer vision tasks, e.g., camera calibration, mensuration, image registration, scene reconstruction, object recognition, and motion analysis (see also Section 1.7 above). Since higher-level descriptions directly depend on the quality of the low-level information, accurate and reliable extraction of these features from images is a central topic. Note, that the original meaning of landmark refers to visually prominent reference ‘points’ in the scene. This includes prominent objects in a particular landscape and also prominent anatomical structures. These landmarks often have been given a name. In addition, landmark also refers to prominent structures in images, particularly 2D and 3D images. Usually salient points in the scene correspond to salient points in the image, but this need not necessarily be the case. In this work, we primarily use the word landmark to denote prominent structures in images.