Multi-Channel Polarization Image Registration Using Improved SURF Based on Phase Congruency and Log-Gabor Filter

Aiming at the difficulties of feature detection and feature matching caused by the geometric distortion and gray-scale radiation difference between dual-channel polarization images, this paper proposes an image registration method based on improved SURF with phase congruency and Log-Gabor filter. Firstly, the phase congruency model is introduced to eliminate the difference in grayscale or contrast between images. The SURF feature points are detected by edge images, which have more stable structural features. Then, the Log-Gabor filter is used to structure a robust local feature description. Finally, the proposed method uses the mutual information measure and RANSAC to perform feature point matching and refinement, and the multi-channel polarization images are accurately matched by the geometrical transformation and bilinear interpolation. The experiment results indicated that the proposed method can effectively improve the matching rate of feature point pairs compared with the classical SURF method. Especially for image registration with different polarization directions and different bands, the proposed method had more accuracy, which can reach within 1 pixel.