The work aims to propose a robust image watermarking algorithm based on IWT-SVD and BRISK to solve the problems of the traditional singular value decomposition algorithm in digital image watermarking, such as diagonal distortion of extracted watermark and the weak performance against combined geometric attacks caused by slow speed of the current feature point matching algorithm. In this scheme, the watermark scrambled by the chaotic map was em-bedded into the singular value matrix of the SVD decomposition after IWT transformation of the host image, and then the BRISK (Binary Robust Invariant Scalable Keypoints) algorithm was used to complete the geometric correction of the distorted image. Finally, according to the extracted watermark image characteristics, the diagonal element correction was performed by the neighborhood averaging method. After the watermark was embedded, the PSNR was higher than 42 dB. Under various conventional signal processing and combined geometric attacks, the normalized correlation values of the extracted watermark and the original watermark were above 0.95, and the diagonal distortion of extracted watermark was significantly improved. Experimental results show that the scheme can increase the matching speed of image feature points, enhance the performance, provide better invisible watermark, and further improve the robustness of image.