An automatic precision registration method based on SIFT and Harris feature for multi-source remote sensing images

Automatic registration of multi-source remote sensing images is a research focus and difficult task. This paper proposes a robust and accurate method for multi-source remote sensing images registration. The proposed method is a two-step process including pre-registration and fine-tuning registration. Firstly, the method detects the matching points by the Scale Invariant Feature Transform (SIFT) algorithm and then the input image is pre-registered by using these points according to polynomial model. As a result, the input image is transformed with the same spatial pixel size and the reference coordinate system as the reference image. Secondly, a large number of feature points are detected based on the Harris corner detector in the input image, tie point pairs are found rapidly by correlation coefficient in a small search window determined in the reference image. Tie point pairs with errors are pruned by Baarda's data snooping method. Finally, both the reference image and the input image are divided into a number of triangular regions by constructing the Triangulated Irregular Network (TIN) based on the selected tie point pairs. For each triangular facet of the TIN, an affine transformation is applied for rectification. Experiments demonstrate that the proposed method achieves precise registration effects.