A robust Type-III data hiding technique against cropping and resizing attacks

Proposes an oblivious information hiding method which allows watermark recovery for signals subjected to cropping and resampling consecutively. We employ multiple embedding of a watermark signal using a Type-III data hiding method which has a better robustness vs. rate trade off than the conventional methods under mean square error distortion measure. Cyclic autocorrelation features of the cropped-resampled signal are used to estimate the amount of cropping. We analytically showed that within a small error range, it is possible to restore the cropped-resampled signal to the cropped signal. Synchronization of watermark detection in the cropped stego-signal is achieved by designing white noise like watermark signals that are uncorrelated with their shifted replicas. For this purpose we use all-pass filters which are orthogonal to all their cyclic shifts. Freedom to hide data is obtained by modulating the phase of the cyclic all-pass filters. Further, we use Reed-Solomon error correcting codes both for introducing redundancy and achieving synchronization. We also address the issue of data hiding under multiple cropping attacks