An algorithm for filtering synthetic-aperture-radar (SAR) interferograms has been developed which substantially improves the performance of subsequent phase-unwrapping. It accomplishes this task by filtering the interferogram strongly in areas where the scene contains little detail, and weakly where the scene contains much detail. The filter is adaptive in that the amount of filtering is scene dependent.
The interferogram is divided into many small, overlapping patches. Each patch is Fourier transformed in two dimensions. The magnitude of the resulting spectrum is then raised to a user-selected power. An exponent of 1 yields no filtering at all; 2 produces heavy filtering. The patches are then transformed back and accumulated through a rectangular, pyramidal window, which reduces edge effects. It is helpful to normalize the total power of each patch in order to prevent radar-bright areas from unduly influencing their neighbors.
The filter is effective in removing residues, the bane of phase-unwrapping, where the phase is not changing rapidly. However, where the phase does change rapidly, or is noisy, no filtering is applied. This effect prevents phase-unwrapping from occurring where the data does not warrant it
This work was done by Charles Werner and Richard Goldstein of Caltech forNASA's Jet Propulsion Laboratory. NPO-20440
