Further Improvements of the Watermarking Scheme

Top

11.1 Diagram Of Proposed Enhanced Watermark System

The diagram of the enhanced watermarking system (encoder and decoder) is illustrated in Figure 1.

Comparing Figure 1 in the current chapter with Figure 1 in chapter 10, we can see that the differences between the enhanced system and the previously proposed one are:

• a.

  For the enhanced encoder, the watermarks are embedded into all wavelet coefficients, regardless of the host signal wavelet power spectrum distribution. In the previously proposed encoder, however, the watermarks are only embedded to the area where the host signal wavelet power spectrum is below the masking thresholds.

• b.

  For the enhanced decoder, the psychoacoustic model is dropped off and replaced by a Savitzky-Golay smoothing filter (William et al., 1992), which is used to de-correlate the host signal and the watermark and decrease the interference from the host signal. Since no psychoacoustic model is needed, the enhanced decoder performs much faster than the previously introduced due to the huge amount saved from the reduction of the masking thresholds estimation.

Top

11.2 Encoder Of The Proposed Enhanced Watermark System

The procedure of the enhanced embedding is similar to the previously proposed, which involves calculating the masking thresholds and spreading the watermark with a PN sequence. Suppose that the data to be embedded are {m_k}, which are normalized wavelet coefficients of the watermark within the range of -1 and 1, for each frame. The embedding rule is:

(11.1) where c_k is the value of the k-th wavelet coefficient in the frame, α is a factor (0≤α≤1) used to control the watermark strength, m_i is the symbol to be embedded in this frame and T is the masking threshold for that frequency sub band. Increasing α typically improves watermarking system robustness by embedding higher energy watermarks. Unlike before, however, α is set in a way that watermarks could be embedded into those above masking thresholds areas without introducing perceptual distortion into the host audio signal.

According to (Swanson et. al., 1998), the introduced distortion is imperceptible as long as its energy is kept below the masking threshold. In our system, we set

(11.2) in order to assure transparent watermark embedding.

Figure 2 to Figure 5 list a typical one frame of audio power spectrum, watermark power spectrum, watermark power spectrum after noise shaping, watermarked audio power spectrum and the masking thresholds.