ABSTRACT
Parallelepiped classification is based on a radiometric model but not on measurement of distance or probability. Every pixel can be represented by its radiometric characteristic on a Cartesian diagram with as many axes as bands. Classification amounts to segmentation of the radiometric scatter diagram. Segmentation can be carried out in several ways. The three-dimensional histogram can be directly segmented making use of only the simplified radiometric model comprising the objects bare soil, chlorophyllian vegetation, water and shadow. The model based on multispectral segmentation is more efficient than the preceding but assumes knowledge of the principal spectral characteristics of objects of the region under study. The model based on radiometric and chorological segmentation is evidently the most efficient, but it assumes that spectral behaviour of major objects of the image is known and that certain chorological laws can be applied. Parallelepiped classification ought to be developed based on a correct radiometric model applicable to the region under study.
