TOWARD A LOW-LEVEL DESCRIPTION OF DOT CLUSTERS - LABELING EDGE, INTERIOR, AND NOISE POINTS

Cluster analysis is an approach to finding structure in raw data patterns. Most of the available algorithms for finding clusters are not generally applicable, however, because their performance varies with the shape of the cluster. This paper presents the first stage of an approach to computing the shapes of dot clusters. It is founded on the general principle that the concept of shape is derived from interactions with the visual world. In particular, a relaxation process for labeling the functional roles that dots appear to be playing in a given arrangement is presented together with many examples of its performance. These labels should provide the primitive descriptions out of which global shape descriptions can be computed. Also, since dot patterns are a simple form of binary image, the relevance of this approach to low-level vision is described. © 1979 Academic Press, Inc. All rights reserved.