This paper presents a novel watermark detection scheme for spread spectrum watermarking (SS) based on the host signal interference cancellation strategy at the watermark detector. The proposed watermark detection scheme is based on the theory of blind source separation (BSS) using independent component analysis (ICA), for watermark detection. As the host signal acts as a noise source, at the blind watermark detector, for SS based watermarking which limits the detection performance of these watermarking schemes. To improve the detection performance of SS watermarking, a new watermark detector based the host signal interference cancellation at the detector using ICA model is proposed. The superiority of watermark detection performance of an ICA based detector over the existing correlation based detector, is based on the fact that an ICA based detector is capable of separating independent components from the received observation (a mixture of underlying independent components i.e. watermarked signal) with negligible host crosstalk residue. Our ICA based detector performs approaches the detection performance of an informed detector. Theoretical analysis and simulations for the real-world data, of the proposed ICA based detector, also reveals this fact. The simulation results for common signal manipulations show that the proposed detector outperforms the conventional correlation based detector. 

Audio clips used for watermarking based on the proposed FSSS based watermarking are listed in Table 1

TABLE 1

SELECTED AUDIO CLIPS

FIDELITY PERFORMANCE

To evaluate the robustness performance of the proposed watermarking scheme is tested against several audio signal degradations. These degradations include addition of white and colored noise, resampling, lossy compression (MPEG Audio compression), filtering, time and frequency scaling, requantization, multiple watermarking, and stirmark benchmark attacks for audio.

·        Addition of White Noise

Figure 1: Decoding Performance, P_e, for an ICA based Detector and a correlation based Detector against AWGN attack with different SNR (dB) values for each selected Watermarked Audio Clip.

·        Addition of Colored Noise

Figure 2: Decoding Performance, P_e against Just Audible Colored Noise Attack on the selected Watermarked Audio Clips.

·        Rescaling

Figure 3: Decoding Performance, P_e, against Time Scaling Attack for ts = ± 1% applied to each Watermarked Audio Clip.

Figure 4: Decoding Performance, P_e, against Frequency Scaling Attack for fs = ± 1% applied to each Watermarked Audio Clip.

·        Resampling

Figure 5: Decoding Performance, P_e, for Resampling Attack; P_e for different values of Resampling Factor is for each watermarked Audio Clip.

·        Requantization

Figure 6: Decoding Performance, P_e, for Requantization Attack applied to each watermarked Audio Clip.

·        Lossy Compression

Figure 7: Decoding Performance, P_e, against Lossy Compression Attack for different Bits Rates using ICA Detector applied to each Watermarked Audio Clip.

·        Filtering

       (a)                                                                            (b)

Figure 9: Decoding Performance, P_e, for Filtering Attack applied to each Watermarked Audio Clip.

(a) Decoding Performance, P_e, for Low-pass Filtering Attack.

(b) Decoding Performance, P_e, for High-pass Filtering Attack.

(c) Decoding Performance, P_e, for Band-pass Filtering Attack.

·        Stirmark Audio Benchmark

TABLE II

AVERAGE DETECTION PERFORMANCE RESULTS ON WATERMARKED AUDIO CLIPS ATTACKED WITH THE STIRMARK AUDIO BENCHMARK.

An improved watermark detector for SS based watermarking is presented in this paper. The proposed detector is capable of canceling the host signal interference at the watermark detector using ICA framework. Bind watermark detection, lower host signal interference at the detector, improved watermarking-rate performance, etc. are the salient features of the proposed ICA based watermark detector. The proposed ICA based detector can be used for SS based watermarking for all types of data, e.g., audio, video, images, etc. Theoretically speaking, detector performance of the proposed ICA based detector is as good as an informed watermark detector. The simulations for real-world data shows that the proposed ICA based detector performs much better than the conventional correlation based detector (commonly used for watermark detection in SS based watermarking). Simulation results against many signal degradations reflect this fact. Moreover, the detection performance of the proposed detector can be improved by employing channel coding.

The subjective audibility-test results, for the proposed audio watermarking scheme, will be included after formal approval (formal subjective testing request is in process).

Currently we are investigating the performance of the proposed ICA based detector for image and video watermarking. We are also looking forward to use the proposed ICA based detector for multimedia fingerprinting for secure distribution on the Web.