Fusion of Multi-variate Edge Detectors for High-resolution Polarimetric Sar Images

Edge detection in SAR images is a difficult problem due to the presence of speckle. However, the statistical properties of speckle in uniform regions of a SAR image can be used for the development of edge detectors. For singlechannel multi-look intensity images, the ratio-detector [1] is widely accepted to be the optimal edge detector. For multi-channel data, it is possible to apply the ratio-detector to each separate channel and fuse the results. Alternatively multi-variate methods can be used. They treat the different channels as a whole and there is no need for subsequent fusion. Furthermore they take the inter-channel correlation into account. We already proposed two edge detectors based on multi-variate statistical hypothesis tests. The first one is based on a test for the difference of variance and applied to SLC images, the second uses a test for the difference of means and is applied to log-intensity images. The two multi-variate edge detectors give complementary results. Hence the idea to fuse these results. Fusion of the results of both detectors for equivalent false alarm thresholds gave poor results. In the article we propose a new method to find the region of optimal fusion for the two edge detectors. The method is based on the combination of two statistical methods for investigating the complementarity of ”experts” and a figure-of-merit for edge detection. Results of applying the proposed method to a high-resolution, polarimetric, L-band E-SAR image are shown.