Digital color image watermarking algorithm based on scale-invariant feature transform(SIFT) and support vector regression(SVR) is proposed in this paper. The input feature vectors are selected in the wavelet domain and then the train model is obtained by applying the support vector regression theory. The watermark information can be embedded or extracted by utilizing the above trained SVR model. The proposed scheme can extract the digital watermark without the help of the original digital image. The scale-invariant feature transform(SIFT) is employed to against the scaling and rotation attacks. To improve the security and robustness, the original watermark is scrambled at first. Experimental results show that the proposed scheme is invisible and robust to common signal processing attacks such as adding noise, JPEG compression, sharpening, smoothing, filtering, contrast enhancement, cropping, rotation, scaling and so on.