Feature-based synchronization correction for multilevel watermarking of medical images

Abstract

Desynchronization attacks cause the global geometric displacements such as Rotation, Scaling, and Translation (RST) of the watermarked image. One of promising way is to embed hiding information in the stable to RST attacks regions defined by feature points. However, a multilevel protection of medical images supposes the embedding of Region Of Interest (ROI), electronic patient record, and fragile watermark. Volume of embedding information is too large to be stored in the restricted regions robust to attacks. First, a preliminary step is required for intensity equalization and detection of the cropped regions. Second, we use a synchronization correction of the global geometric attacks with the following compensation of possible local warping. The applied frequency transform such as digital wavelet transform or digital shearlet transform with single value decomposition is chosen due to a changeable volume of embedded information. Arnold’s transform is additional procedure increasing a security. Our experiments show that the proposed synchronization correction provides accuracy of the extracted watermarks up to 92%.