With the increase of the Synthetic Aperture Radar (SAR) sensor resolution, a more detailed analysis and a finer
description of SAR images are needed. Nevertheless, when dealing with urban areas, the high diversity of manmade
structures combined with the complexity of the scattering processes makes the analysis and information
extraction, from high resolution SAR images over such areas, not easily reachable.
In general, an automatic full understanding of the scene requires the capability to identify both relevant and
reliable signatures (called also features), depending on variable image acquisition geometry, arbitrary objects
poses and configurations. Then, since SAR images are formed, by coherently adding the scattered radiations
from the components of the illuminated scene objects, we can make the assumption that, the SAR image is a
superposition of different sources. Following this approach, one alternative for a better understanding of the HR
SAR scenes, could be a combination between the Principal Components Analysis (PCA) and the Independent
Components Analysis (ICA) decompositions. Indeed, while the PCA exploits at most the information stored in
the sample covariance matrix, the ICA is a de-mixing process whose goal is to express a set of random variables
as linear combinations of statistically independent component variables. Such an approach could be useful for
the recognition of urban structures, in HR SAR images. In this paper, we compare the Principal Components
(PCs) to the Independent Components (ICs). Furthermore, we present some preliminary results on learning and
decomposing SAR images, using PCA and ICA.