Modifications in the SIFT operator for effective SAR image matching

Abstract

With the increasing availability and rapidly improving the spatial resolution of synthetic aperture radar (SAR) images from the latest and future satellites like TerraSAR-X and TanDEM-X, their applicability in remote sensing applications is set to be paramount. Considering challenges in the field of point feature-based multisensor/multimodal SAR image matching/registration and advancements in the field of computer vision, we extend the applicability of the scale invariant feature transform (SIFT) operator for SAR images. In this article, we have analysed the feature detection, identification and matching steps of the original SIFT processing chain. We implement steps to counter the speckle influence, which deteriorates the SIFT operator performance for SAR images. In feature identification, we evaluate different local gradient estimating techniques and highlight the fact that giving up the SIFT's rotation invariance characteristic increases the potential number of matches when the multiple SAR images from different sensors have been acquired with the same geometrical acquisition parameters. In the feature matching stage, we propose to assist the standard SIFT matching scheme to utilise the SIFT operator capability for effective results in challenging SAR image matching scenarios. The results obtained for SAR images acquired by different sensors using different incidence angles and orbiting directions over both rural and semi urban land cover, highlight the SIFT operator's capability for point feature matching in SAR imagery.

Robust Feature Matching Method for SAR and Optical Images by Using Gaussian-Gamma-Shaped Bi-Windows-Based Descriptor and Geometric Constraint

“…Figure 1a shows that a lot of noise was detected as interest points falsely by the DoG detector. Figure 1b shows the result of the improved DoG detector adopted in [14][15][16]. The blue and green partially enlarged details show that false detection still exists and that many real feature points are overlooked, respectively.…”

“…However, it also leads to a reduction in the accuracy of feature localization because all features originate from the smoothed scale-space images. The researchers proposed another approach [15] based on the previous work in [14] by suppressing the orientation computation in feature detection and assigning a uniform orientation for all features on the basis of the observation that the orientation assignment affects the robustness of the descriptor. In order to improve the performance further, local geometric is added to constrain the matching procedure [16].…”

Comparison of Oil Spill Classifications Using Fully and Compact Polarimetric SAR Images

“…Ellipticity χ (CP) 9 CPD Standard Deviation (CP) 10 Conformity Coefficient (π/2) * Features 1-9 are extracted from dual and compact polarimetric SAR data following the methods introduced in Section 2.4, while Feature 10 is only available for π/2 mode. "CP" stands for features derived from compact polarimetric SAR data in order to distinguish them from those calculated from quad-pol SAR data.…”

“…Therefore, satellite synthetic aperture radar (SAR) data from ERS-1/2 (European Remote Sensing Satellites), ENVISAT (Environmental Satellite), ALOS (Advanced Land Observing Satellite), RADARSAT-1/2 and TerraSAR-X have been widely used to detect and monitor oil spills [1][2][3][4][5][6][7][8] due to the large spatial coverage, all-weather conditions and imaging capability during day-night times [9]. In addition, airborne SAR sensors, such as Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) developed by JPL at L-band and E-SAR (developed by DLR), have proven their potential for scientific research on ocean or land [10,11].…”

Modified Ransac for Sift-Based Insar Image Registration

“…When applied to image registration, Random Sample Consensus (RANSAC) [14] is often used with SIFT to remove outliers (mismatched pairs of points). Although a lot of image registration results by SIFT and modified versions with RANSAC are reported [15][16][17][18][19][20], few works have been done on InSAR image registration and little attention has been paid to RANSAC. Furthermore, the "maximization of inliers" criterion of the original RANSAC is not optimal for InSAR image registration for the number of residues (NOR) [21] of the interferogram obtained through this criterion is not the fewest.…”