The appropriate parameter retrieval algorithm for feature-based SAR image registration

摘要: This paper is dedicated to investigate the appropriate parameter retrieval algorithm for feature-based synthetic aperture radar (SAR) image registration. The widely-used random sample consensus (RANSAC) is observed to be instable for its inappropriate estimation strategy and loss function for SAR images. In order to enable a stable and robust registration for SAR, an extended fast least trimmed squares (EF-LTS) is proposed which conducts the registration by least squares fitting at least half of the correspondences to minimize the squared polynomial residuals instead of fitting the minimal sampling set to maximize the cardinality of the consensus set as RANSAC. Experiment on interferometric SAR image pair demonstrates that the proposed algorithm behaves very stably and the obtained registration is averagely better than that by RANSAC in terms of cross-correlation and spectral SNR. By this algorithm, a stable estimation for any kind of 2D polynomial warp model with high robustness and accuracy can be efficiently achieved. Thus EF-LTS is more appropriate for SAR image registration. �2012 SPIE.