Performance Evaluation of Unsupervised Coregistration Algorithms for Multitemporal SAR Images

In this paper, we present three algorithms for the multitemporal synthetic aperture radar (SAR) images coregistration. The proposed algorithms are a 2-D cross correlation, a 1-D parabolic based, and a 2-D projective transformation. The 2-D cross correlation algorithm is used to obtain coarse estimation of the displacement for coregistration. In the second method, two independent 1-D parabolic interpolations are calculated to refine the estimation of the peak location of the cross correlation matrix with subpixel accuracy. Finally, in the third method, a 2-D projective transformation is employed to align the SAR images using point correspondences and the cubic interpolation. The performance evaluation of these algorithms are provided based on the coherence magnitude and the absolute displacement error for a point target using a corner reflector in the scene. The experimental results obtained on real recorded multitemporal satellite SAR data demonstrate the effectiveness and the computational complexity of these algorithms.