AUTOMATED MULTI-SOURCE REMOTE SENSING IMAGE REGISTRATION BASED ON PHASE CONGRUENCY

Abstract

Automated registration of multi-source remote sensing data, e.g., optical, SAR, LiDAR data, could be difficult due to their heterogeneity. This paper proposes a method based on the concept of phase congruency, a measure of feature significance robust to change in illumination and contrast. We first calculate the phase congruencies respectively for the input image and the reference image to reduce local illumination and contrast difference caused by the heterogeneity or radiometric variation. Minimum moment of phase congruency is used to select feature points in the input image, followed by a normalized cross correlation to determine their correspondences in the reference image. Prior georeferencing information and image pyramid guide the above matching process, which is then refined by means of least squares matching (LSM) method. The proposed method uses random sample consensus (RANSAC) to remove the large correspondence errors caused by e.g., shadow and shielding. Finally, image registration is achieved by determining a projective transformation between the two images based on the final set of correspondences. The proposed method is evaluated with multi-source remote sensing images, including optical images, SAR images and LiDAR data. The registration quality is evaluated by using manually collected check points. The results demonstrate that the proposed method is robust and can achieve a registration accuracy which can be comparable to that produced by manual registration.