Application of an Improved SIFT algorithm in GPR images

Classical scale invariant feature transform (SIFT) and its corresponding improved algorithm are widely used in computer vision and remote sensing images. However, in the detection and matching task of GPR images, the final feature matching quality of traditional SIFT algorithm will be directly affected by the richness of underground information. This paper introduces a more robust method to adapt to GPR images. The method combines the classical SIFT and NCC, and finally selects the best matching pair by stable matching strategy, and extracts more correct matching points and improves the accuracy rate from GPR images with low pixel density and less feature information. The experimental results show that, compared with the classical SIFT algorithm and the conventional combination of RANSAC and sift algorithm, the proposed algorithm can obtain more correct matching point pairs, and can be effectively applied to the detection of underground environment changes by GPR tomography images.