A new affine invariant descriptor framework in shearlets domain for SAR image multiscale registration

To cope with the problem of accurate synthetic aperture radar (SAR) image registration, a new affine invariant descriptor framework in shearlets domain is proposed. This framework consists of three steps in sequence. First, a new affine invariant descriptor is developed based on scale invariant features transform (SIFT) and kernel space theory, which is called kernel affine invariant SIFT (KA-SIFT). Then the new descriptors are used to match the feature points detected from the different sub-images in the corresponding layer, which are obtained by the shearlet decomposition and affine-SIFT (ASIFT) algorithm. Finally, a coarse-to-fine procedure is adopted for gradual optimizing transformation parameters to achieve the multiscale registration. Experimental results show that this framework is more robust and accurate than some state-of-the-art methods. It can accurately detect the changes of the reservoirs and lakes before and after the Wenchuan earthquake, validating the proposed framework.