Saliency Detector for SAR Images Based on Pattern Recurrence
Saliency Detector for SAR Images Based on Pattern Recurrence
复制标题
DOI：
10.1109/jstars.2016.2521709
复制
发表时间：
2016-02
影响因子：
5.5
通讯作者：
Haipeng Wang;F. Xu;Shanshan Chen
中科院分区：
文献类型：
--
作者：
Haipeng Wang;F. Xu;Shanshan Chen
In order to detect targets from nonuniform background in synthetic aperture radar (SAR) images, this paper proposes a new approach for saliency detection based on the idea of pattern recurrence. The pattern recurrence quantifies the saliency of a local patch by how well it can be reconstructed from patches in the background. The similarity between polarimetric SAR (PolSAR) patches is defined as the likelihood that both patches belong to the same Wishart distribution. A simple saliency indicator is defined as the normalized variance of the nonlocal similarity map. To detect target of different sizes, multiscale patches are also used. The proposed approach is first tested on optical images and results are compared with other saliency detection method and experiments of eye fixation prediction. Test results on real SAR images show that the proposed method has robust performance on target detection with and without the presence complicated background.
