Discrete-object modeling of remotely sensed scenes

Abstract

Remotely sensed scenes can be modeled as collections of discrete, three-dimensional objects that cast shadows on a background. Approaching scenes from this perspective has led to two related lines of research. First is the geometric/optical modeling of a forest canopy, in which conifers are modeled as cones whose size and spacing vary according to functions established by field measurements. This canopy model is 'L-resolution' in nature - the objects are smaller than the resolution cells of the image and cannot be resolved individually. Second is scene modeling in which image variance is taken as a function of the relationship between the size, shape and spacing of objects and the resolution cell size of the digital image derived from the scene. This modeling is 'H-resolution' in nature - the objects in the scene are assumed to be larger than the resolution cells of the image, and thus can be individually distinguished. Both approaches illustrate the utility of the discrete-object scene model in extracting information from remotely sensed scenes.