Key points are not available for this paper at this time.
Uniformly distributed feature extraction and availability of a sufficient number of correctly matched points between the input images are the key challenges for remote sensing optical image registration. Because of its robustness and distinctiveness, the scale-invariant feature transform (SIFT) is a well-known approach for an automatic image registration. However, the features obtained from the SIFT algorithm are not uniformly distributed, and sometimes, the number of matched features is insufficient to provide subpixel accuracy in the registration of remote sensing optical images. In this letter, a modified version of SIFT is proposed to obtain uniformly distributed matched features. Then, the bivariate histogram and the random sample consensus have been used to refine the initial matches. Finally, a reliable matching criterion is proposed to maximize the number of matches.
Paul et al. (Thu,) studied this question.