Automatic Relative Radiometric Normalization Algorithm Based on Pseudo-Invariant Neighborhood

An automatic relative radiometric normalization algorithm of multi-temporal remote sensing images based on pseudo-invariant neighborhood was introduced in this paper. The main purpose of this research was to develop a robust algorithm of relative radiometric normalization to minimize imaging differences of multi-temporal satellite images. The main idea was to construct statistical regression model of relative radiometric normalization by pseudo-invariant neighborhoods, which were obtained by extracting steady ground point correspondences. The algorithm's detailed processes were as follows: First, a method of image matching was applied to recognize ground point correspondences of multi-temporal remote sensing images. Second, steady ground point correspondences were determined by matching results. Third, sample dataset of steady ground point correspondences was obtained by pseudo-invariant neighborhoods. Finally, a statistical regression model of relative radiometric normalization was constructed to calibrate imaging differences of multi-temporal remote sensing images. By the experiments, we could see that the total RSME of reference image and calibrated image was reduced apparently in comparison with the total RSME of reference image and un-calibrated image. We could conclude that this modified algorithm was more accurate and efficient than traditional algorithm, and the result of calibration was more objective and dependable.