Distinguishable zero-watermarking scheme with similarity-based retrieval for digital rights Management of Fundus Image

Abstract

Zero-watermarking scheme can provide durable and distortion-free digital rights management (DRM) for fundus image which plays an important role in diagnosis of ocular diseases. However, existing zero-watermarking schemes probably identify a similar fundus image as a copy, because they rarely consider the distinguishability for image. In addition, when the number of fundus images is large, it is difficult to obtain corresponding ownership shares accurately for copyright identification, because there is no retrieval mechanism in these schemes. To address these issues, a distinguishable zero-watermarking scheme which fuses similarity-based retrieval is proposed for DRM of fundus image. In our proposed scheme, distinguishable and robust features of fundus images are extracted based on the gray-scale variation. The ownership shares are constructed using visual secret sharing (VSS) by combining watermark and the master shares generated from these features. Once a suspected fundus image is found, the similarity-based retrieval is performed to retrieve the corresponding ownership share based on the feature of suspected image. After that, the copyright is identified by stacking the master share of suspected image and the retrieved ownership share. Experimental results on three public databases demonstrate that 1) Ownership shares corresponding to specific fundus images can be retrieved precisely. When fixing the false positive rate to 0.001, the mean false negative rates are not higher than 0.0693. 2) Copyrights of fundus images can be identified accurately and reliably. The mean bit error rates of recovered watermark are not higher than 0.0460.