| Summary: | In this study, a novel coarse-to-fine image registration algorithm based on feature matching using the Gabor filters is provided for repeat-pass SAR interferometry (InSAR). In the coarse registration, we propose a feature-based algorithm developed using the Gabor odd filter ratio-based operator (GOFRO), and the innovative feature based on the GOFRO (GOFRO-F) consists of the improved nonlinear scale space and the descriptor. The establishment of nonlinear scale spaces informed by the GOFRO enhances the stability of feature detection. Also, an improved 64-dimensional descriptor based on the GOFRO is presented to optimize the capability of feature matching. In the fine registration, we employ the normalized correlation coefficient algorithm (NCCA) and the random consensus sample (RANSAC) for determining offsets and the sub-pixel projection model calculation. The datasets of interferograms are derived from the 0.3 THz stepped-frequency radar. Compared to other algorithms, registration experiments of both simple and complex scatters show that GOFRO-F is more suitable for high precision image registration because of its enhanced robustness to the speckle noise. Furthermore, the three-dimensional (3D) reconstruction is conducted based on our proposed algorithm. The results demonstrate that our proposed algorithm facilitates the attainment of millimeter-level spatial resolutions for InSAR in THz band (THz InSAR).
|
|---|