A mean based algorithm for the multi-temporal SAR image filtering

Any despeckling method for SAR images is a trade-off between speckle reduction level and preservation of image resolution. The Time-Space Filter (TSF) proposed by Colt$80uc et al. (2000), despeckles the multi-temporal SAR images while preserving the radar reflectivity, the textures and the edge contrast. The price paid for this feature integrity is a non-uniform despeckling: the speckle is reduced only in the homogenous areas, whereas the edge vicinities and the textured areas remain noisy. We propose a new version of the TSF, the Mean based Time-Space Filter (M-TSF), which reduces everywhere the speckle, disregarding the image content. The residual noise in the images filtered by M-TSF remains multiplicative and stationary. This latter property makes M-TSF appropriate for pre-treatment in automatic edge detection. A comparison between M-TSF and the Lee et al. (1991) filter for multipolarization SAR images leads to an interesting conclusion: M-TSF is a frequential version of the Lee et al. filter.