Speckle Filtering of Polarimetric SAR Interferometry Data
4555 Overlook Ave., SW,
Washington DC, 20375-5351,
(2) AEL Consultants, Cupar,Fife, KY15 5AA, Scotland, United Kingdom
(3) DLR German Aerospace Center, DLR, D-82230, Oberphaffenhofen, Germany
Polarimetric SAR interferometry has generated great
interest for forest applications. The complex
interferometric coherence from combinations of
polarizations is computed to form the 6x6 coherency matrix.
This matrix is generally filtered by a boxcar filter to
reduce speckle, and then the linear combination of
polarizations is optimized producing three optimized
coherences and their associated interferometric phases. In
earlier work, the maximum phase difference between these
three interferometric phases was used to estimate tree
heights. Recently, a coherent mixture model of a random
volume and ground proposed by Papathanassiou and Cloude 
has been applied to extract tree heights and the extinction
coefficient based on interferometric coherence for
polarimetric SAR. In both approaches, the accurate
estimation of coherence is critical for tree height
estimation. Interferometric coherence is a statistical
average. It requires averaging many samples from the same
distribution. It has been shown that an insufficient number
of samples would produce an overestimate of the true
coherence. For example, for 1-look interferometric SAR
data, the coherence is at the maximum value of 1. The
second problem associated with coherence computation is
that the coherence is underestimated when averaging samples
from difference distributions. Thus, the commonly used
boxcar filter would produce erroneous coherences especially
near forest boundaries, or in inhomogeneous vegetated
areas. For examples, at forest edges, a 5x5 window would
contain samples from two distinct distributions. The
indiscriminate average produces a lower coherence at the
edges as dark lines (lower coherence) of 3 to 5 pixel
wide.. Most forest areas are not homogeneous, such as those
in the Glen Affric Project. Better speckle averaging than
the boxcar filter is needed. The basic principle in speckle
filtering of polarimetric interferometric data is in the
selection of pixels of the same scattering characteristics
to be included in the average. Polarimetric SAR has the
capability of characterizing the scattering mechanism of a
medium. The filtering is done on the 6x6 coherency matrix
based on the speckle filter that minimize the mean square
error . The filtered image successfully corrects the
deficiency of the boxcar filter. The problem of dark (low
coherence) lines along forest edges is eliminated.
Importance in the tree height extractions using the Glen
Affric data will be demonstrated.
 K.P. Papathanassiou,
S.R. Cloude and A. Reigbor, "Single and Multi-Baseline
Polarimetric SAR Interferometry over Forested Terrain,"
Proceedings of EUSAR 2000, pp123-126, Munich, Germany.
J.S. Lee et al., "Polarimetric SAR Speckle Filtering and
Its Implication for Classification," IEEE TGRS, vol.37, no.
5, September 1999.