An efficient SAR edge detector with a lower false positive rate

An efficient ratio-based edge detector (RBED) is proposed for synthetic aperture radar (SAR) images to decrease the false positive detection rate. Low false rate is achieved by taking full advantage of the integrative information associated with multiple adjacent edge-pixels. The RBED introduces a novel 2D separable filter. The 2D filter is represented as the direct product between a 1D vertical filter and a 1D parallel filter. The vertical filter is heuristically designed as the product of a power function and a Gaussian function, simply called the power-modulated-Gaussian function with a weaker tail. Under the non-ideal edge model, this power-modulated-Gaussian vertical filter keeps the false positive rate low and provides a good localization precision. The parallel filter is set as a flat bell-shaped function that is designed by taking a good trade-off between the Gaussian function and the gate function. Such a flat bell-shaped parallel filter with two flexible parameters can be properly adjusted so that the false positive rate can further be decreased. The constant false alarm rate and localization bias of the RBED are analysed theoretically. A set of objective and subjective experiments demonstrate that the RBED can markedly decrease the false positive rate and obtain closed skeleton edges.