Multiscale salient region detection and salient map generation for synthetic aperture radar image

Abstract

Saliency, which refers to distinctive parts of an image that can immediately attract attention without prior information, is significantly meaningful in machine vision, image interpretation, and moreover in quickly extracting vehicle targets or man–made constructions in remote sensing image processing. Different from optical images, synthetic aperture radar (SAR) image has unique characters such as multiplicative speckle noise, local intensity variation, and no color information. However, it is still a challenging work for SAR images to detect salient region and to generate saliency map with the methods performed on optical images. To address this problem, this paper presents a stable salient region detection and salient map generation method for SAR image. We propose a new local complexity metric, which is insensitive to speckle noise and can effectively describe the local intensity variation of SAR image. In addition, via incorporating a stable distribution distance measure, the self-dissimilarity metric is redefined. Using these two components, we construct the saliency metric and generate the salient map. Experimental results demonstrate the accuracy, robustness, and stability of our method for SAR images.