Multi-look technique for SAR speckle filtering

An synthetic aperture radar (SAR) speckle filtering method was developed using principal component analysis to analyze sub-aperture images for radar cross section (RCS) reconstruction. A parameter vector is defined to describe a unit point in the scene. Then, the covariance matrix of the vector is decomposed into orthogonal signal and noise subspaces. The energy concentration of the matrix eigenvalues is then used to evaluate whether the environment of the current point is homogeneous or heterogeneous. The variable part of the parameter vector is projected onto the signal subspace to identify intrinsic structural features of the scene and estimate the RCS. Test results show that the method compares favorably with other de-speckling methods, with good preservation of edge and fine object detail and reasonably speckle inhibition degree. Since the method does not used fixed PDF models, the method can be used with various sorts of SAR.