Image Features and Landmarks
Local image features, like protuberances, ridges, and edges, are often detected at early stages of image processing and are used at higher stages to arrive at compact descriptions of targets, for example, for identifying the anatomical structures of a patient captured in a given image. Many local image features are defined for general purposes: edges, for example, are the locations at which the brightness changes sharply and are detected for a variety of purposes. Many different features have been proposed for higher-level image processing, e.g., image segmentation, image recognition, and for landmark detection. In many medical image processing programs, anatomical landmarks are detected first to determine the locations of the voxels in given images. Only after detecting the landmarks can the body sizes and shapes of the patients be normalized, allowing introduction of a coordinate system appropriate for using computational models. For the landmark detection, local image features that are specifically observed at each landmark location must be designated. This may pose some difficulty if corresponding landmarks in different patients have different appearances and because multiple locations can have appearances similar to those of some landmarks.