Robustness against geometric attacks is one of the most important issues in digital watermarking. A novel geometric robust watermarking scheme that uses computer-generated holograms as the watermark is presented. To maintain imperceptibility and robustness, a quantization embedding algorithm is adopted to embed the mark hologram into the low-frequency subband of the wavelet-transformed host image. In the detection process, the geometric distorted watermarked images are recovered first by the proposed improved geometric correction method, which is based on the scale invariant feature transform, the invariant centroid, and the pulse coupled neural network. Then the mark holograms are extracted from the recovered images. In comparison with the traditional geometric estimation method, the suggested improved geometric correction method can estimate the geometric distortion parameters more accurately and needs less auxiliary information. Compared with other watermark schemes using digital holograms, the proposed method has the distinct advantage of robustness to geometric attacks. The experimental results demonstrate that the proposed method has good robustness to resist geometric attacks and common attacks including rotation, scaling, translation, image flipping, combined attacks, filtering, occlusion, cropping, and JPEG compression.