In this paper, we analyze the performance of scalar quantization based data hiding techniques with decreasing cover-signal sizes under mean squared error distortion measure. We introduce a new scheme, called multiple codebook data hiding, that enables conventional embedding/detection techniques to utilize the permitted embedding distortion more efficiently. The proposed method treats the embedding distortion introduced to a cover-signal as a random variable, and utilizes the fact that decreasing cover-signal size increases the deviation of the embedding distortion from its expected value. This is exploited by embedding the watermark into a variant of the cover-signal that yields a lower embedding distortion. In the proposed method, variants of the cover-signal are obtained by deploying a set of real unitary transformations known to both embedder and detector. For the given cover-signal, the embedder chooses a transformation basis and embeds the message in the transformed cover-signal, whereas the detector has to search all transformations of the received signal for the embedded message. We evaluate the performance improvement due to multiple codebook data hiding and compare it with the conventional (single codebook) approaches, under additive white Gaussian noise attacks, in terms of the bound on the probability of detection error. Performance results obtained from simulation and by applying the technique to image watermarking problem under JPEG compression attack are also presented.
All Science Journal Classification (ASJC) codes
- Control and Systems Engineering
- Signal Processing
- Computer Vision and Pattern Recognition
- Electrical and Electronic Engineering
- Data hiding
- Embedding distortion