Analysis and Evaluation of Speckle Filters by Using Polarimetric Synthetic Aperture Radar Data Through Local Statistics

Speckle noise, a granular noise, occurs in synthetic aperture radar (SAR) data due to the interference of reflected signals with several scatterers in a resolution cell. One of the simplest techniques to suppress speckle noise from polarimetric SAR data is to use local statistics. The Lee filter employs sample mean and variance of pixels of data degraded by speckle noise, which can be multiplicative, additive, or a mixture of both, in a searching window. The refined Lee method utilizes directional windows with local minimum mean square error (LMMSE), which ensures superior maintenance of spatial resolution and features. The Lee-Sigma filter is constructed on the basis of the two-sigma probability, which effectively suppresses the speckle noise. However, insufficiencies were observed in the generation of a biased evaluation, blurring of edges, and suppression of point targets. To eliminate these insufficiencies, the improved Lee-Sigma filter was developed, which employs the minimum mean square error estimate as a priori mean. Thus, an excellent maintenance of point targets and subtle details is displayed. In this study, the Lee, refined Lee, Lee-Sigma, and improved Lee-Sigma filters were evaluated using full polarimetric data. The evaluated results indicated that the improved Lee-Sigma filter performed better than other local statistics filters.