A family of fuzzy edge detectors for SAR images

Abstract

Segmentation is an important technique for information extraction out of Synthetic Aperture Radar (SAR) imagery. In most cases, it is difficult to achieve substantial accuracy, due to noise and ambiguities caused by SAR processing and/or SAR geometry. Detection of edges is one approach to SAR image segmentation. The aforementioned phenomena have the same impact to edge detectors applied on SAR images. False edges are created and actual edges are lost due to noise. This imposes the use of thresholds connected with false alarm probabilities. A new approach for edge detection in SAR images, which overcomes the necessity of choosing some arbitrary thresholds is described in this paper. The Fuzzy Edge Detector (FED) takes into account the uncertainties that occur in SAR images. It does not only discriminate between edge pixels and no-edge pixels, but assigns each image pixel a degree of being an edge pixel. Additionally, the possible edge directions in each pixel are estimated. This intermediate fuzzy output enables a very flexible post-processing and thus an efficient edge detection processing chain. The resulting edge map is still a fuzzy map. It can be directly used to improve classification of SAR images. Defuzzification can be performed by the use of certain thresholds. This leads to crisp edge maps. In this case, false alarm probabilities are to be considered.