<title>Image Segmentation Using Prototype Similarity</title>

Abstract

An image segmentation technique using prototype similarity is described here. The prototype similarity is a method for transforming an attribute image into a set of symbols, each of which represents the relationship of a local region to other parts of the image. It consists of two main steps: (1) prototype generation and (2) inference. Generating prototypes is equivalent to finding a maximal set of mutually dissimilar regions using a similarity relation. A similarity relation is a symmetric, reflexive binary relation and not an equivalence relation. It is not bound by metric properties. The generated set of prototypes is used to transform the attribute image into a symbolic image. Some a priori information about the scene is used to infer meaning of each cell in this symbolic image. The segmentation results of this technique on FLIR images are included.