Vector filtering of single-look complex SAR data based on adaptively weighted local order statistics

A new vector filter is proposed, that outputs a complex pixel value belonging to a local processing window as a weighted vector median. Unlike scalar median applied to detected images, vector median yields detail-preserving yet unbiased estimates of reflectivity. The weights of neighboring vectors are adaptively calculated from local statistics of the complex data so as to reflect the ability of a pixel vector to fit the vectors lying in the local processing neighborhood in the weighted least squares sense. The design of weights avoids smearing structured edges and at the same time allows recovery of targets. Experimental results on SLC ERS-1 SAR images demonstrate that the weighted vector approach outperforms local statistics (scalar) processing of the detected image, especially in terms of sharpness retention and preservation of point targets and region boundaries.