In the process of network transmission, when an exception occurs (such as forgery, tampering, information confusion), digital medical image, as a diagnostic basis, can not serve as the evidence of medical accident case. And the ROI of medical image is unable to tolerate significant changes. In order to deal these problems, we have proposed a multiple watermarks algorithm that uses Arnold scrambling to preprocess the original multiple watermarks, improving the security of watermarking, and combining the visual feature vector of image with the encryption technology and the concept of third-party. Moreover, the sophisticated process is needless to find the Region of Interest (ROI) of medical images. So compared with the existing medical watermarking techniques, it can embed much more data, with less complexity. The experimental results show that the scheme has strong robustness against common attacks and geometric attacks.