An Image Description for Object Definition, Based on Extremal Regions in the Stack

Abstract

A promising image description is produced by dividing an image into nested light spots and dark spots by considering the image simultaneously at many levels of resolution [Koenderink, 1984]. These spots each include an image extremum and are thus called extremal regions. The nesting can be specified by a tree indicating the containment relationships of the extremal regions. This tree description, with each region described by intensity information, size, shape, and most significantly a measure of the importance, or scaleM, of the spot, absolutely and relative to its containing spot, ought to be usable in finding meaningful image objects when it is used together with a priori information about the expected structure of the image or its objects. This paper will describe work in the development of a computer program to compute such a description and its application to the display and segmentation of images from x-ray computed tomography and nuclear medicine.