A MMSE SPECKLE FILTER FOR FULL RESOLUTION SAR POLARIMETRIC DATA

The usual polarimetric speckle filters optimally combine the polarizat ion channels into a single image [10] or only restore the radiometric information [8], i.e., only the four I(hh), I(vv), I(hv) and I(vh) bac kscattered intensities (or the three I(hh), I(vv), I(hv) intensities i n the reciprocal case). The phase differences for one-look images axe not restored. This implies a loss of information compared to the initi al data, which contain five independent real parameters plus one absol ute phase for the one-look scattering matrix format in the reciprocal case. In this paper we develop a fully polarimetric minimum mean squar e error (MMSE) filter under the multiplicative speckle noise model ass umption for one-look images. This model assumes the data to be the pro duct of the non-stationary ''unspeckled'' signal by statistically inde pendent speckle noise, characterized by the covariance matrix of a sta tionary complex multivariate Gaussian random process, representing the sensor effects. For each pixel, one obtains on output of the filterin g process a complex ''unspeckled'' scattering matrix and in addition, the complex degrees of coherence between the polarization channels. Fu rthermore, an adaptive window of sufficient size, defined by the coeff icient of variation and various contrast ratios computed in an improve d span image, allows the detection of structural features, ensuring re liable statistics for input parameters. Hence, speckle is highly reduc ed and the spatial resolution is not degraded. The extra information b rought up by the filter (phase differences and magnitude of degrees of coherence) will be illustrated in this paper on NASA/JPL data.