Traitor-tracing (aka fingerprinting) has received much attention as a possible solution for protecting media copyrights. However, current schemes for image and video fingerprinting lack robustness against geometric attacks. We propose a novel semi-blind fingerprinting scheme that can cope with such attacks. The scheme improves a state-of-the-art high-rate fingerprinting code that can resist tens of colluders and Gaussian noise but has no resistance against geometric attacks. Our scheme uses compressed SURF (Speeded-Up Robust Features) image features as side information in order to estimate and invert any geometric attack in a given class. We consider simple linear attacks (affine transforms), and more complex ones (homography and image warping). Our Estimation-Elimination algorithm estimates the attack parameters by matching image features and eliminating iteratively suspected outliers. We also compare this method to an adapted version of RANSAC (Random Sample Consensus).The fingerprints are embedded securely and invisibly using Spread Transform Dither Modulation (STDM) applied to the intermediate level of a Laplace decomposition of the image. The fingerprints are robust against common attacks such as averaging, interleaving, addition of Gaussian noise, JPEG compression (with quality factor Q=45), cropping (50% of the image area), affine transforms, homography and image warping.
【 预 览 】
附件列表
Files
Size
Format
View
A feature-based fingerprinting scheme robust to desynchronization
PDF
download
878987797
028-85220240
