Shadow Detection in SAR Images: An OTSU- and CFAR-Based Method

Abstract

Shadow has gradually become an important feature in current synthetic aperture radar interpretation, while shadow detection itself seemingly didn't attract enough attention in the past research. This paper contributes to propose a new method for shadow detection in SAR images. Our algorithm is divided into three stages: firstly, suspected shadow areas are extracted from the input SAR image with a dual-threshold OTSU based segmentation; secondly, an improved two parameter-constant false alarm rate method is utilized to detect objects; at last, we design a discrimination strategy to remove false areas from suspected shadows, and then the left regions are final shadow detection results. Experiments based on images from MSTAR dataset present that our method comprehensively outperforms another published algorithm, WD-CFAR, which demonstrates the feasibility of applying our algorithm to practical SAR shadow detection tasks.