Multiresolution Feature Encodings

Multiresolution data structures require that information be summarized or encoded as the resolution decreases. Pyramids and quadtrees provide convenient structures for representing this process, especially for intensity information in images. Encodings are also possible for other image features, such as lines and curves, and objects that are compact or elongated. Methods are described for representing such features at successively lower resolutions, and for using the resulting structures to find objects or regions with particular properties. Advantages of the techniques include (1) their processing speed, if suitable parallel hardware is available; (2) the fact that they provide approximate representations of the features at a wide range of scales; (3) the fact that they condense global information about features into local labels, which can then interact with other such information, and can also provide feedback to influence the condensation processes themselves.