Ground resolution elements seen by a multispectral sensor, may in some cases, consist of mixtures of object categories so that many of the pixels of an image are not characteristic of any category or class. It would be desirable to have a model to represent the combination categories in terms of simpler homogeneous ones and then estimate the corresponding proportions. A method of correlation of intensity variations, may reveal if a given image is representable by a linear mixture model. However, the intensity variation may be due to effects as shadows or variations in the incident light conditions. It is shown that these effects would also produce a linear variation and experiments for several LANDSAT images show that this is the case. Consequently, the estimation problem is difficult to solve.
False color displays may filter this kind of linear variations producing uniform combined colors for homogeneous areas. The reason for the above conclusions is that, statistically speaking, the intensity variation from pixel to pixel of an homogeneous image, has the same sign for all spectral channels.
Date of this Version