Abstract
A representation of image intensity structure, the multiscale orientation field (MOF), is described, and its application to biomedical image segmentation is discussed. The MOF is composed of orientation vectors at every pixel and at multiple scales. The vectors are computed from the outputs of an artificial visual system composed of spatial filters whose design is described. An abstract feature space for orientation measurement is defined, and a filter sensitivity function that yields a desirable mapping into the feature space is described.<>