NRDSP: A novel assessment of SAR image despeckling

Abstract

In order to properly characterize the preservation of structural features for synthetic aperture radar (SAR) image despeckling, a novel metric called the NRDSP (no-reference despeckling structure-preserving) is put forward and investigated. To begin with, the DSSIM (distance for structural similarity) metric is presented to characterize the distance between the ratio image and the despeckled result. As an improvement of the ENLR (equivalent number of looks of ratio images), the NENLR (nominal number of looks-oriented ENLR) metric is proposed with the advantage of rationally bounded function. The DSSIM and NENLR are indispensable and complementary for characterizing the degree of structure-preserving. Furthermore, based on the new factors of DSSIM and NENLR, the novel NRDSP metric is proposed to appropriately measure the structure-preserving performance. Lastly, we carry through some testing for simulated and real SAR images, and it is verified by the observers’ evaluation as well as scientific data that the proposed NRDSP metric is more consistent with the perceptual perceptions than other metrics, and possesses the better effect for characterizing structure-preserving performance.