The work aims to improve watermarking algorithm robustness against geometric attack and propose a new watermarking algorithm against geometric attack based on SIFT geometric correction. First, the feature points of the original image and attacked image were respectively extracted with SIFT. Before watermark extraction, the feature points of the original image and attacked image were matched and the geometric correction of the attacked image was carried out based on the matched feature points. In the embedding process, the robustness of the watermark image was enhanced in combination with the stability of the singular value decomposition (SVD) eigenvalue and the linear independence of non-negative matrix factorization (NMF). The algorithm in the paper was robust in shear, JPEG, noise and other attacks, and the NC value of the extracted watermark was higher than 0.98. The NC value of watermark image under attack of RST could also reach over 0.97. The proposed algorithm against geometric attack can resist all kinds of geometric attacks effectively and also is robust.