Pattern Recognition Letters 13 (1992) 463-470 June 1992 North-Holland Visibility-related image features Leila De Floriani Istiluto per la Matematica Applicma, Consiglio Nazionale delle Ricerche. Genova, ltal.v Philippe Jeanne David Sarnoff Research Center, Subsidiary of SRI International, Princeton, NJ 08543-5300, USA George Nagy Rensselaer Polytechnic Institute, Electrical. Computer and Systems Engineering Department, Troy, NY 12181, USA Received 28 January 1991 Abstract De FIoriani, L., P. Jeanne and G. Nagy, Visibility-related image features, Pattern Recognition Letters 13 (1992) 463-470. The intervisibility of Iwo points in a digital grey-scale image array is defined by analogy belween tile inlensily values of the picture elements and surface elevation on topographic surfaces. The distributions of the pah'wise visibility (a boolean function) between image poinls, of the visibilily count of Ihe number of points visible from each poitlt, and of the distam'e q/'lhe horizon from each poillt in selecled direclions, reveal sigrfil'icant image characleristics. These features allow the classification of images according Io type, and subregions according to corltent and erwironme,'ll. The discriminating power of these fealures is augmented by the introduction of an observalion heigh! that allows shifting emphasis from local to global characteristics. I. Introduction Many types of generic features have been pro- posed for classifying the content of digitized grey- scale image arrays. Among the most widely used are edge detection, zero-crossings, shape-specific matched filters, medial axis transforms, Fourier, Haar, and other spatial frequency transforms, cor- relation and co-occurrence matrices (Haralick, 1979), and morphological operations (Serra, 1982). An image can be conceived as a topographic surface where the elevation represents the image Correspondence to: P. Jeanne, SAGEM, Centre d'Eragny, UR96, Avenue du Gros Chine, Eragny, B.P. 51, 95612 Cergy- Pontoise Cedex, France. brightness (see Figure 1). This analogy has been developed in particular for locating features such as peaks and valleys (hot spots and glints), pits and valleys (dark spots or deep shadows), and cliffs (edges) (Watson et ai., 1985). Others have also taken advantage of the analogy between grey-scale and elevation (Toriwaki and Fukumura, 1978; Beucher, 1982), but not, to our knowledge, with regard to visibility. The visible region of a point on a topographic surface corresponds to the por- tions of this surface one can see from that point. This concept is well-defined and is useful in a variety of applications such as geomorphology, navigation, or terrain exploration (Shapira, 1990). Our approach uses visibility relations which de- pend both on local and global characteristics of the 016%8655/92/$05.00 ,~, 1992 -- Elsevier Science Publishers B.V. All rights reserved 463