Multiresolution Edge Detection in SAR Images

The presence of a strong multiplicative noise known as speckle complicates edge detection in Synthetic Aperture Radar (SAR) images and makes usual edge detectors inefficient. However, speckle is very well modelled statistically, and we here use this property to develop two non-linear multiresolution edge detectors for SAR images. A normalized ratio of local averages on opposite sides of the central pixel is computed in different directions and on different resolution levels for every pixel position. The pixelwise maximum or mean value over all scales yields an edge strength map, from which the edges are finally extracted. The difference between our approach and wavelet-based edge detection is briefly discussed, some statistical properties are examined, and the improvement compared to edge detectors operating on one single scale is shown on simulated SAR data.