A Coarse-to-Refined Matching Method for Multisensor Remote Sensing Image Registration

Multisensor image registration is necessary in many applications of remote sensing imagery, the crucial problem is how to establish the correspondence between the features extracted from the reference and input image. Generally, most existing methods only use feature similarity or intensity similarity. In this paper, a coarse-to-refined method, which combines modified scale invariant feature transform (SIFT) feature similarity in coarse matching and cluster reward algorithm (CRA) in refined matching, is developed. To achieve refined registration, two transformation models are used. The experimental results demonstrate that the proposed method is effective and achieves subpixel registration accuracy.