In this paper we analyze the potentialities of two different representations commonly used for polarimetric SAR images, for segmentation and classification purposes. The first representation is based on the covariance matrix of the considered polarimetric channels, while the second is based on the "H-A-alpha" decomposition. For both representations, we devise a segmentation + classification scheme, and we evaluate their performance on a set of C-and L- band polarimetric AirSAR images. On the basis of the achieved results, we devise a hybrid scheme, that uses a segmentation algorithm based on the full covariance matrix, then jointly classifies the pixels of the identified homogeneous regions considering the "H-A-alpha" representation. Results show that the hybrid scheme provides a significant performance improvement with respect to the approaches based on a single representation.
[1]
Pierfrancesco Lombardo,et al.
Optimum model-based segmentation techniques for multifrequency polarimetric SAR images of urban areas
,
2003,
IEEE Trans. Geosci. Remote. Sens..
[2]
T. M. Pellizzeri,et al.
Classification of polarimetric SAR images of suburban areas using joint annealed segmentation and “H/A/α” polarimetric decomposition
,
2003
.
[3]
Pierfrancesco Lombardo,et al.
Segmentation-based joint classification of SAR and optical images
,
2002
.
[4]
Eric Pottier,et al.
A review of target decomposition theorems in radar polarimetry
,
1996,
IEEE Trans. Geosci. Remote. Sens..