Relative radiometric normalization of SAR images based on bi-direction linear regression model

The Relative radiometric normalization (RRN) of multi-temporal synthetic aperture radar (SAR) images is very important for change detection. Many RRN techniques have been developed up to now. Among them, the No-Change set regression normalization algorithm (NC algorithm) is proved to have the best performance in optical image processing field. However, as our paper shows, the presence of speckle noise in SAR image will make the estimations of RRN parameters biased in NC algorithm. In order to deal with this problem, we propose a RRN method suitable for SAR images. In the approach, the biases of RRN parameters estimations are removed by a bi direction linear regression model. And a recursive weighted least square method is included to improve robustness. The effectiveness of the proposed approach is confirmed with experimental results compared to NC algorithm.