An adaptive blind watermarking algorithm based on the human visual system (HVS) and support vector regression (SVR) in wavelet domain is proposed in this paper. Thanks to the good non-linear function learning and generalization capability, SVR is adopted to model the HVS to determine the adaptive watermark strength by considering the each image features which affect the visibility of the watermark. Compared with other conventional algorithms, the advantages of this algorithm are that it can increase each watermark strength by considering the HVS characteristics, simultaneously keep it imperceptible, so that the robustness of watermark is enhanced. Extensive experiments show that our technique possesses significant robustness to be against the various attacks.