.NTA2MQ.MjU4MTEz

CONTRAST FEATURES EXTRACTION F. MEYER Centre de Morphologie Mathématique, Ecole des Mines 25 Rue Saint-Honora 77305 Fontainebleau - France. ABSTRACT By generating a binary mask of the contrasted structures in a grey tone image, the algorithms presented in this paper lead to a fine characterization of chromatin patterns. / INTRODUCTION Most texture analysers work only on binary images, though the real images have many grey levels, and they use thresholding to pass from the second to the first. This method is well suited for many problems. However, some images make sense only if they are considered as grey-tone functions : we can, for example, mention the chromatin patterns in the nucleus. In a first section we describe the existing ways to characterize this kind of image. In a second section we shall explain how we can generate a binary mask for the gra¬ dient and the contrast peaks. A third section describes some measurements on these features and their application to the characterization of cancerous chromatin patterns. EXISTING METHODS A very good survey of these methods has been made by R.M. HARALICK under the title "Automatic Remote Senser Processing" in Digital Picture Analysis (Réf. 1). He clas- sifies them in two catégories. The first includes ail méthodes which characterize the texture of an image by a measure, or a number. The second type of methods trans forms images into other images.