Robust Multi-Source Image Registration for Optical Satellite Based on Phase Information

Robust image registration is a vital challenging task, especially for multi-source satellite images that may have significant different illumination. A coarse-to-fine registration algorithm based on phase information is proposed. The coarse registration is implemented using Fourier-polar transform and phase correction based on phase congruency. The fine registration is first implemented by dividing large images into blocks. Then, the uniformly distributed corners and Principal Phase Congruency (PPC) images are extracted. After that, the corresponding points of extracted corners are obtained based on Phase Correlation of Principal Phase Congruency (PCPPC), followed by a new outlier removal method. Experiment results revealed the robustness, accuracy, and distribution quality less than 1.0 of the proposed algorithm. The matching correct rate is about 94.7 percent for Data Set 2 due to considerable topography variations, and more than 96.6 percent for data set 4 despite significant different or inverse intensity, which can reach 100 percent with our outlier removal method.