Shadow Segmentation in SAS and SAR Using Bayesian Elastic Contours

Abstract

We present a variational framework for naturally incorporating prior shape knowledge in guidance of active contours for boundary extraction in images. This framework is especially suitable for images collected outside the visible spectrum, where boundary estimation is difficult due to low contrast, low resolution, and presence of noise and clutter. Accordingly, we illustrate this approach using the segmentation of synthetic aperture sonar (SAS) and synthetic aperture radar (SAR) images. The shadows produced from these imaging modalities often times offer more consistent pixel values with clearer contrast to the background than the targets pixels themselves, and thus we focus on the extraction of shadow boundaries rather than target boundaries. Since shadow shapes can vary under approximately affine transformation with different target range and aspect angle, we incorporate an affine-invariant, elastic shape prior based on the shape analysis techniques developed in [2] to the active contour model. We show experimental results on both a simulated SAS and a simulated SAR image database in three segmentation scenarios: without shape prior, with similarity-invariant shape prior, and with affine-invariant shape prior.