SAR Image Edge Detection from Multichannel Data

The main goal of building edge detection is to obtain a map of man-made structure edges of the investigated scene. Different detectors have been developed exploiting synthetic aperture radar (SAR) data, based on the use of the reflectivity difference (working with SAR amplitude images) or of the phase difference (working with SAR interferometric images) between neighboring pixels. In this thesis a jointly approach using both the amplitudes and the interferometric phase of two complex SAR images will be discussed. This approach is based on the hypothesis that information related to building edges can be retrieved in the two different data domains. The technique is based on stochastic estimation theory, using mainly Markov random fields (MRF). Compared to classical amplitude-based edge detectors and to phase-based ones, the join approach exhibits an improvement in terms of detection accuracy, false alarm rate, and building shape recovery. The edge detection provides a map of the boundaries of man-made structures of the area under investigation. In general the building edge detection happens in the optical domain using gradient-based techniques such as Sobel filter. With the new high resolution of the Synthetic Aperture Radar (SAR) systems, the edge detection has become a task in the radar domain as well. Unfortunately, gradient-based techniques cannot be directly applied to SAR data, because of the presence of the speckle effect which has a multiplicative nature. A despeckling filter could be readily applied, but as consequence there is a loss of resolution. But there are mainly two approaches which are able to take into account the speckle without resolution loss: One approach is based on radio techniques for the edge detection while the second approach is based on stochastic methods which depend on Markov Random Field.
During the last years, several InSAR approaches have been using for the edge detection, based on the idea that the edges of building could be found from the height information which is related to the interferometric phase. But in general InSAR approached require detailed a priori information on the observed scene: some approach, for example, requires the detection of the shadowing areas.
Otherwise there is a stochastic approach modeling the image as a Gaussian MRF (GMRF) with local hyperparameters which does not require any a priori information about the scene or assumptions on building shape. The local hyperparameters are seen as an indicator of the spatial correlation of the pixels. The detection of building edges is found through the estimation of the local hyperparameters. Since approaches based on amplitude or phase SAR data provide effective results, using techniques based on the joint employment of complex SAR data (amplitude and interferometric phase) could improve the edge detection both in term of false alarm rate and detection probability and to better recover the building shapes. In urban areas, each couple of neighboring pixels involved in a building edge is characterized by both a big reflectivity discontinuity (in SAR amplitude images) and a big phase discontinuity (in InSAR images).