Filtering of spatially invariant image sequences with multiple desired processes

The selection of a filter vector for linear filtering of a sequence of spatially invariant images of an object or scene to maximize the ratio of desired component energy to undesired component and noise energy in the filtered image is considered. The filtered image is a weighted linear combination of the images of the sequence, and the filter vector is the set of weights. New results extend this filtering technique to spatially invariant imaging applications with multiple desired components. The filter vector which provides the filtered image requires the solution of an eta -dimensional eigenvector problem, where eta is the number of desired processes in the image sequence. An explicit result is given for eta =2. Special results are also given for the cases in which the filter is designed only to suppress undesired processes or only noise. A four-image multispectral example illustrates the method.<
>