Audio Watermarking in DCT: Embedding Strategy and Algorithm

1. Audio Watermarking in DCT: Embedding Strategy and Algorithm Signal Processing, 2008. ICSP 2008. 9th International Conference on Institute of Information Science, Beijing Jiaotong University, Beijing, P. R. China • Gaorong Zeng andZhengding Qiu 1

2. Outline • Introduction • Background • Proposed algorithm • Experimental results • Conclusion • Reference 2

3. Introduction • Internet application widely makes it easier for music pirates to copy and distribute music and songs around the world illegally. • Compared with watermarking in the time domain, Watermarking in the transform domain becomes more feasible because of better robustness. • Audio watermarking based on DCT is one of many common techniques. • Many strategies were proposed to implement the watermarking such as modifying low frequency, middle frequency, or high frequency DCT coefficient of an audio signal. • This paper presented a method of modifying high frequency DCT coefficient for increasing the robustness of audio watermarking. 3

4. Background • DCT is short for Discrete Cosine Transform • Like the discrete Fourier transform, a DCT operates on a function at a finite number of discrete data points. 4

5. Background(cont.) • DCT coefficients noise model and related work • Total noise energy is: • Noise sensitivity coefficient: (Defined by Ma Yiping) 5

6. Background(cont.) • β (i) is a constant number • Obviously, we can’t select the valid location based on β (i) , because the contribution of every DCT coefficient position is equivalent. 6

7. Background(cont.) • SNR and PSNR can be calculated as followed: • We can see that the SNR and PSNR are identical when modification quantity E(i) is fixed • In the paper, we choose the DC coefficient as the embedding position because it is the most important component with the largest amplitude. 7

8. Proposed algorithm(embedding) • Step1: The original audio is divided into two families of frames in term of energy, high energy section and low energy section. • Step2: For high energy section, decomposing every frame into non-overlapping 8/L blocks and the DCT is performed. Choosing the DC coefficients F(0) as the embedding positions. • Step3: Embedding is based on QIM method. 8

9. Proposed algorithm(extracting) • Step1: Perform the same procedure as (1) and (2) of the embedding phase and extract the DC coefficient. • Step2: Judge the embedded information Ṽ(0) of each block by • Step3: Repeat Steps (1) to (2) until all the watermark bits are extracted. • Step4: Revert the one-dimension bit series Ṽ (k) ( 0 ≤ k < M1 ×M2 ) to the two-dimension image information 9

10. Experimental Results • Comparing three methods(LSB, LSB&DCT, Bit4&DCT) by using PSNR 10

11. Experimental Results(cont.) • Comparing three methods(LSB, LSB&DCT, Bit4&DCT) by using PSNR 11

12. Experimental Results(cont.) • Comparing three methods(LSB, LSB&DCT, Bit4&DCT) by using PSNR 12

13. Conclusion • Choosing the DC coefficient with the largest amplitude as the embedding position through the SNR and PSNR are identical when modification quantity E(i) is fixed. • Experimental results show watermarking is robust to common signal processing operation and mp3 compressing, and audibility is satisfaction. • This paper presented an audio watermarking method inserting data in DC coefficient instead of inserting in AC, the concept of this paper gives me another point of view of steganography using DCT. 13

14. Reference • Ma Yiping, Han Jiqin, “Audio Watermark In DCT Domain: Strategy And Algorithm” [J], Chinese Journal Electronics, 2006;34(7): 1260-1264. • I. J. Cox, J. Kilian, T. Leighton, T Shamoon.“Secure spread spectrum watermarking for multimedia”[J], IEEE Transaction on Image Processing,1997,6(12):1673-1687. • Liu Tong, QIU Zhengding, A quantization-based image watermarking algorithm[J], Journal of China institute of Communications, 2002,23(10):89-93. 14