ROBUST AUDIO WATERMARKING USING FREQUENCY-SELECTIVE SPREAD SPECTRUM
Hafiz Malik, Ashfaq Khokhar, Rashid Ansari
Dept. of Electrical and Computer Engineering University of Illinois at Chicago, Illinois, USA
ABSTRACT
This paper presents a novel scheme for embedding digital watermark in audio signals. The proposed scheme employs frequency selective spread spectrum (FSSS) analysis for embedding watermark into a fraction of the audible frequency rang. Most of the existing spread spectrum (SS) based watermarking schemes utilize complete audible frequency range for watermark embedding. The proposed FSSS scheme exploits the frequency masking characteristics of the human auditory system (HAS) to ensure the fidelity and robustness of the embedded information. The robustness performance of the proposed FSSS scheme is evaluated against intentional and unintentional attacks that include random chopping, resampling, requantization, lossy compression, lowpass, highpass and bandpass filtering, addition of colored and colored noise, time and frequency scaling, multiple watermark embedding, and stirmark benchmark attacks for audio. High watermark embedding capacity, reduced host interference at the watermark detector, low embedding distortion, secure embedding, and improved multiple watermark embedding capability are the salient features of the proposed watermarking scheme.
SIMULATION RESULTS
Audio clips used for watermarking based on the proposed FSSS based watermarking are listed in Table 1
TABLE 1
SELECTED AUDIO CLIPS
| Singer Name /Song Title | Type | Duration (Sec) | |
| 1 | Backstreet Boys, I Want It That Way … | Pop, (Pop1) | 22 |
| 2 | Lata Mangeshkar, Kuch Na Kaho … | Melodic, (Melodic) | 15 |
| 3 | Asha Bhosle, and Richa Sharma, Kahin Aag Laga … | Pop, (Pop2) | 10 |
| 4 | Nusrat F. A. Khan, Afreen Afreen … | Classical, (Classical) | 20 |
| 5 | Suzanne Vega, Tom's diner | Female Vocal, (Vocal) | 5 |
FIDELITY PERFORMANCE
| Original Audio Clips | Pop1 | Melodic | Pop2 | Classical | Vocal |
| FSSS Based Watermarked Audio Clips | Pop1 | Melodic | Pop2 | Classical | Vocal |
ROBUSTNESS PERFORMANCE TEST
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
· Addition of Colored Noise
Figure 2: Decoding Performance, Pe against Just Audible Colored Noise Attack on the selected Watermarked Audio Clips.
· Rescaling
Figure 3: Decoding Performance, Pe, against Time Scaling Attack for ts = ± 1% applied to each Watermarked Audio Clip.
Figure 4: Decoding Performance, Pe, against Frequency Scaling Attack for fs = ± 1% applied to each Watermarked Audio Clip.
· Resampling
· Requantization
Figure 6: Decoding Performance, Pe, for Requantization Attack applied to each watermarked Audio Clip.
· Lossy Compression
Figure 7: Decoding Performance, Pe, against
Lossy Compression Attack for different Bits Rates using ICA Detector applied
to each Watermarked Audio Clip.
Figure 8: Decoding Performance, Pe, against Lossy Compression Attack for different Bits Rates using Correlation Detector applied to each Watermarked Audio Clip.
· Filtering
(a) (b)
(c)
Figure 9:
Decoding Performance, Pe,
for Filtering Attack applied to each Watermarked Audio Clip.
(a) Decoding
Performance, Pe, for
Low-pass Filtering Attack.
(b) Decoding
Performance, Pe, for
High-pass Filtering Attack.
(c) Decoding
Performance, Pe, for
Band-pass Filtering Attack.
· Stirmark Audio Benchmark
TABLE II
AVERAGE DETECTION PERFORMANCE RESULTS ON WATERMARKED AUDIO CLIPS ATTACKED WITH THE STIRMARK AUDIO BENCHMARK.
|
Stirmark Attack |
Pe_ICA |
Pe_Correlation |
Stirmark Attack |
Pe_ICA |
Pe_Correlation |
|
addbrumm_100 |
0 |
0.2258 |
exchange |
0 |
0.2258 |
|
addbrumm_1100 |
0 |
0.2258 |
extrastereo_30 |
0 |
0.2258 |
|
addbrumm_2100 |
0 |
0.2258 |
extrastereo_50 |
0 |
0.2258 |
|
addbrumm_3100 |
0 |
0.2258 |
extrastereo_70 |
0 |
0.2258 |
|
addbrumm_4100 |
0 |
0.2581 |
fft_hlpass |
0.0323 |
0.2258 |
|
addbrumm_5100 |
0 |
0.2581 |
fft_invert |
0 |
0.2258 |
|
addbrumm_6100 |
0 |
0.2581 |
fft_real_reverse |
0 |
0.2258 |
|
addbrumm_7100 |
0.0323 |
0.2903 |
fft_stat1 |
0.1931 |
0.4839 |
|
addbrumm_8100 |
0.0323 |
0.3226 |
fft_test |
0.1931 |
0.4839 |
|
addbrumm_9100 |
0.0323 |
0.3226 |
flippsample |
0.1613 |
0.4839 |
|
addbrumm_10100 |
0.0646 |
0.3548 |
invert |
0 |
0.2258 |
|
addnoise_100 |
0 |
0.2258 |
lsbzero |
0 |
0.2258 |
|
addnoise_300 |
0 |
0.2258 |
normalize |
0 |
0.2258 |
|
addnoise_500 |
0 |
0.2258 |
rc_highpass |
0.0323 |
0.2258 |
|
addnoise_700 |
0 |
0.2258 |
rc_lowpass |
0 |
0.2258 |
|
addnoise_900 |
0 |
0.2258 |
smooth |
0 |
0.2258 |
|
addsinus |
0 |
0.2258 |
smooth2 |
0 |
0.2581 |
|
amplify |
0 |
0.2258 |
stat1 |
0 |
0.2258 |
|
compressor |
0 |
0.2581 |
stat2 |
0 |
0.2258 |
|
dynnoise |
0 |
0.2581 |
zerocross |
0 |
0.2258 |
|
echo |
0.0323 |
0.3548 |
zeroremove |
0.0323 |
0.2258 |
CONCLUSION
A novel spread spectrum based audio watermarking scheme is presented in this paper, the proposed scheme inherits the salient features of conventional SS based watermarking schemes. In addition, the FSSS based watermarking has higher embedding capacity, introduces lower embedding distortion, and more secure embedding than the existing SS based watermarking. Simulation results show that the proposed scheme is robust to the common intentional and unintentional watermarking attacks if ICA based detector is used for watermark detection process. The detection performance of the correlation based detector can be improved by employing channel coding. The subjective audibility testing results for the proposed audio watermarking scheme will be included after formal approval (the formal subjective testing request is in process).