Fast and Automatic Registration Method for Large Multi-spectral Remote Sensing Images

Aiming at the registration of large multi-spectral remote sensing images,a fast and automatic registration method based on feature points was proposed.Because the size of the multi-spectral remote sensing image is quite large,and the amount of calculation is also very great,the theory of feature grid based on grey value,entropy and uniformity principle was proposed.Feature grids chosen from a two-degree regular mesh are calculated in the subsequent process to reduce the calculation.Meanwhile,the theory provides condition for detecting SIFT(Scale Invariant Feature Transform) feature points parallelly,and for improving the primary feature matching step,so the efficiency and accuracy are increased.The proposed method was applied to remote sensing images taken by CBERS-02B.The experimental results with large remote sensing images clearly indicate that the proposed approach can achieve sub-pixel precision,decrease the runtime of the process,and the requirement of large remote sensing image process is satisfied.