The use of spline based wavelet filtering to improve classification processing of SAR imagery

SAR imagery poses a significantly more difficult problem in classification than optical imagery because of the coherent scattering process and the geometric morphology of the scattering object. The resolution of the image and the impact of the speckle can cause significant confusion in the classification processing. The tendency of the wavelength to be on the order of the scattering geometry generates morphological characteristics in the imagery that, although may contain information, tends to confuse classification algorithms. The difficulty in these imagery is that all of these process are coupled. Conventional image processing techniques do not tend to separate these behaviors. Wavelet processing techniques offer some capability in these cases. The process is local so that any filtering can be confined to a specific image local, the process is directional, and the results of any processing of wavelet coefficients can be incorporated in the reconstructed image. In this work the authors use the cubic spline based wavelet developed by Chui (1992). These wavelet descriptors have special characteristics that allow high performance low pass and band pass filters to be constructed. These filters have very good sidelobe performance characteristics. The wavelet decomposition and reconstruction also have efficient computational implementations.