RETRIEVING MULTISPECTRAL SATELLITE IMAGES USING PHYSICS-BASED INVARIANT REPRESENTATIONS

We present a set of algorithms and a search strategy for the robust co ntent-based retrieval of multispectral satellite images. Since the pro perty of interest in these images is usually the physical characterist ics of ground cover, we use representations and methods that are invar iant to illumination and atmospheric conditions. The representations a nd algorithms are derived for this application from a physical model f or the formation of multispectral satellite images. The use of several representations and algorithms is necessary to interpret the diversit y of physical and geometric structure in these images. Algorithms are used that exploit multispectral distributions, multispectral spatial s tructure, and labeled classes. The performance of the system is demons trated on a large set of multispectral satellite images taken over dif ferent areas of the United States under different illumination and atm ospheric conditions.