Multimodal Remote Sensing Image Registration Based on Local Phase Congruency and Keypoint Filtering

Robust registration is able to register the complementary information captured by multimodal images. Owning to the nonlinear intensity differences and repeating patterns in multimodal images, many classic descriptors are unable to achieve good performance. Aiming at this problem, a remote sensing image registration method based on key point description and filtering is proposed. First, this work utilizes a new keypoint descriptor with Log-Gabor filters to characterize the oriented phase congruency information in multimodal images. Then, a keypoint filtering is performed on all descriptors by mean-shift clustering to remove the repeating patterns. This approach is evaluated on three commonly used data sets, which are used for structure map construction and image registration comparisons. The experimental results show that the proposed method is able to preserve highly similar structure features and remove most repeating keypoints in multimodal images, which indicates this method is efficient and can achieve good registration performance.