Research of Accurate Registration Method for Multispectral Remote Sensing Images

Aiming at the inconsistent registration between different spectral image within a group of multi-spectral remote sensing images, a fast and automatic registration method based on feature points was proposed in this paper. On the base of image entropy, an annular moving window is used to find the region of interest automatically and quickly, and the Scale Invariant Feature Transform (SIFT) algorithm is used to extract features. To increase the accuracy, we improve the primary feature matching step, and use cosine theorem and space distance constraint condition to eliminate false matching points, then extract the most stable feature points to calculate transform parameters to complete registration. At last, the mutual information of before and after the registration and the root mean square error (RMSE) of feature points are used to measure the extent of the registration. By many experiments on a large number of multi-spectral images acquired by the CBERS-02B satellite, the result indicates the method can achieve sub-pixel registration accuracy for each spectral image, and can decrease the runtime of process.