River Segmentation Based on Visual Salience Calculation of Spectral Residual and Region Growing in SAR Images

Abstract

In the problem of the SAR image river segmentation, the threshold method and the region growing method are two widely used segmentation methods based on water pixel information. In view of the fact that the precision of the segmentation result for the threshold method is low and the segmentation result of the region growing method has voids which lead to the problem of the high missing alarm; for the first time, this paper introduces the visual salience detection theories into the SAR image river segmentation and presents a method that combines visual salience calculation of spectral residual and region growing. This method first binarizes the preprocessed SAR image, then extracts the saliency map of the binary image by utilizing the spectral residual model, and finally segments the river region by using the region growing method on the saliency map. Compared with the threshold method, the region growing method and the method that combines the threshold method and the region growing method, the experiments demonstrate that the method proposed by this paper has much better precision and comprehensive segmentation performance. Apart from that, it also effectively solves the problem of the voids existing in the segmentation results for the traditional region growing method. Therefore, the method presented by this paper can be applied to the SAR image river segmentation in the applications such as water resources planning and flood disaster prevention.