An adaptive threshod segmentation algorithm to extract dark targets from SAR images

Abstract

An adaptive threshold segmentation algorithm to extract dark targets from SAR images is presented, which is the key procedure to establish an automatic oil spill detection system. The extracted dark targets will then be sent to a classifier, such as a neural network, to discriminate oil spills and look-alikes. In order to reduce the calculation amount of following classifier, some simple filters are applied to reduce the look-alikes as many as possible while ensuring all oil spills are remained. Accurate local background estimation is required to determine the dark targets. Usually, the mean brightness of a small sliding window is used to estimate the background brightness. But it is not suitable for big dark targets. To avoid the effect of big dark targets, the proposed algorithm firstly estimates a rough background and then iteratively refines the background estimation. In each step, the rough dark targets are extracted based on the rough background. The new background is then calculated by removing the dark targets. The procedures above repeat iteratively and finally the best estimated background and dark targets are obtained simultaneously. The first guess background can be calculated by fitting the azimuthal averaged brightness trend along the range with parabolic curve.