Robust image processing for remote sensing data

Remote sensing has become an important resource for numerous areas of application. Efficient methods for analysis and visualization of this data are needed as new satellites with improved capabilities are planned and constructed. This paper describes numerical and imaging techniques that are based on a statistically robust singular value decomposition (RSVD). This algorithm characterizes the main features of remotely sensed data without the distorting and masking effects due to the presence of relatively rare subpopulations. Ancillary numerical tools associated with this algorithm can be used for the identification and visualization of the rare subpopulations. Problems discussed include a brief description of RSVD, storage of its output, image processing, and modifications to allow efficient processing of high-dimensional remote sensing data. Examples of the application of these methods to a SPOT data set are presented