A new edge detection technique for SAR images with speckle

Edge detection in synthetic aperture radar (SAR) images is made more difficult by the presence of multiplicative noise, known as speckle. The technique presented in this paper compensates for the speckle, while detecting edges and other features, by using the known speckle statistics. It also has other advantages over conventional edge filters. Traditional edge detection filters look for differentials, either between or within fixed areas, centered on each pixel of the image. The technique described takes a different approach, by using the speckle statistics to define a homogeneous area around each pixel. The size of this typically irregular area defines the “connectedness” of each pixel, and is the output of this process. Pixels connected to large neighbourhoods usually belong to a smooth area, while those with few connections indicate edges or other features. This technique is insensitive to orientation or curvature in a feature, and sets the window size and shape adaptively, based on local conditions. Also, the data is not pre-filtered to remove speckle, as is generally required with other edge filters. A RADARSAT image is used to evaluate this new technique, in comparison with the commonly used Sobel edge filter, and results are presented