STEGANOGRAPHIC COMMUNICATION CHANNEL ... - IEEE Xplore

12th International Conference on Mathematical Methods in Electromagnetic Theory June 29 – July 02, 2008, Odesa, Ukraine

STEGANOGRAPHIC COMMUNICATION CHANNEL USING AUDIO SIGNALS S. Hernández-Garay, R. Vázquez-Medina, L. Niño de Rivera and V. Ponomaryov National Polytechnic Institute, ESIME Culhuacan Graduate School, Av. Santa Ana 1000, 04430, San Francisco Culhuacan, D.F., México [email protected] Abstract – Using MP3 and WAV audio digital signals an algorithm to build a steganographic communication channel is presented. This algorithm is conformed by the insertion and extraction procedures and it use direct sequence spread spectrum (DSSS) to insert confidential (subliminal) information in MP3 and WAV audio digital signals. Evaluation tests were applied to the different steganographic communications channels. According with the obtained results this algorithm can be used to property or exploitation right protection of the audio digital signals.

I. INTRODUCTION The use of steganographic techniques was reconsidered in protection systems in 1996, when the low bit coding, phase coding and spread spectrum techniques was studied by W. Bender [1] to hiding data into digital medias with minimal degradation. From the year 2000 several research papers have arisen related to this subject; in 2001 the compression resistance of MP3 files with hiding data using spread spectrum, time domain insertion and phase domain insertion techniques was studied by L. Gang [2]. In 2003, an insertion procedure modifying the signal amplitude and phase [3], and a power low insertion technique [4] were proposed by Kaliapan. And again he presents in 2005 a technique based on spectral domain of the audio signal [5]. In 2006, some insertion methods that use spread spectrum techniques were proposed by H. Matsuoka [6]. Finally, one of the works more mentioned in this area was developed by Kaliappan in 2007 [7]. In that work, hiding techniques of the audio signal using a truncated series of the first expansion coefficients of the Bessel function are presented. There are many techniques to build steganographic communication channel [1]. In this work, the spread spectrum techniques are used and then the subliminal information will be distributed by all spectral components on the cover digital audio signal. In this case, some pseudorandom sequence independent of the digital audio signal (cover signal) and of the subliminal information data sequence must be used. In both ends of the communication channel, the pseudo noise sequences must be the same sequence and they must be synchronized. III. ALGORITHM DESCRIPTION In this section the algorithm to build steganographic communications channels using the spread spectrum techniques is described: 1) The cover signal of duration t secs. must be restricted to the interval [-1, 1], it must be sampled using the sampling rate tm in order to have a total of M= ttm samples, and it must be segmented in R=(ttm)/k signal segments with length k. 2) A binary pseudo noise sequence W, should be produced with length k and binary values {-1, 1}. 3) The insertion procedure of the subliminal information should be realized using DSSS, using the equation (1),

Sj

A j DWb j ; b j ^ 1,1` ; 1d j d m 1 b j ® ^ ` D A W 1 ; b 0 , 1 j j ¯

(1)

where Aj is the j segment of the cover signal, bj is the j binary digit of the bit stream on the subliminal information, m is the size in bits of the subliminal information, W is a pseudo noise sequence with uniform statistical distribution and length k, Į is the attenuation factor. This factor permits the imperceptibility of the subliminal information and Sj is the j segment of the steganographic signal, which contains the hiding bit, bj. This procedure adds like noise the subliminal information, bit-to-bit, in the cover signal. The expected error

978-1-4244-2284-5/08/$25.00 © 2008 IEEE

427


function between cover and steganographic signals must be approximately the one that is indicated by the equation (2):

DW 1 b j ; b j ^ 1,1` 1 b j ® ¯DW 1 ; b j ^0,1`

Error j

(2)

4) The recovery of each bit inserted must be calculated by means of the following equation:

~ Sj

>

@

E S jW ; 1 d j d m

~

(3)

~

If S j !0 means that the hiding bit is “l”, and if S j 0 means that the hiding bit is “0”. Considering that the channel capacity is given in terms of the mutual information, according equation (5):

C

~ ~ Max I S ; S d I S ; S

^ `

(4)

Also considering, that C is measurable and it is given by the amount of bits that can be inserted in the cover

~

signal, the conditional entropy between S and S called “equivocation” can be calculated. In addition, knowing that the equivocation is limited by the entropy of S , then the inequality (5) can be expressed:

~ H S C d H S / S d H S

(5)

IV. ALGORITHM EVALUATION A. Insertion and extraction test. To evaluate the algorithm proposed 12 MP3 files were considered. The duration of each MP3 file is t=20 sec, which are sampled at tm=44100 samples/sec. In order to evaluate the quality of the insertion and extraction procedures, 3 different kind of musical sort: “rock”, “pop”, and “ranchera” were considered. The evaluation criteria were the following ones: (a) the cover signal must be not degraded, (b) the steganographic signal should be listened equal to the cover signal, (c) the subliminal information should be imperceptible, and (d) the cover signal should have capacity of support the size of the subliminal information (payload). For the insertion procedure, each cover signal was divided in segments of length k=1024 samples; and then R=875 useful segments were obtained. The payload was of 144 segments, because the size in bits of the subliminal information is m=144. To evaluate the degradation level the Signal to Noise Relation (SNR) and the Mean Square Error (MSE) were used, which can be expressed by the equations (3) and (4), respectively and the obtained results are shown in the table I for different cover signals and different Į values.

SNR dB

MSE MP3 Files Rock.mp3 Pop.mp3 ranchera.mp3

ª ¦ M A2j º j 1 « » 10 log10 « ¦ mk A j S j 2 » ¬ j1 ¼ M 1 Aj S j 2 ¦ j 1 mk

Recovery Percentage 100% 97.4% 100% 92% 100% 95%

Į 0.0009 0.0002 0.009 0.002 0.0008 0.0006

Table I. SNR and MSE to different cover audio signal and different Į values.

978-1-4244-2284-5/08/$25.00 © 2008 IEEE

(3)

428

(4) SNR(dB) 38.4032 51.467 23.1593 36.223 26.0273 48.526

MSE 81e-6 39e-7 81e-4 4e-5 64e-4 359e-6


For this example, the relative difference between the cover and steganographic signals and its Fourier Transform are shown in the Fig. 1 using an attenuation factor D=0.0009.

(a)

Fig. 1.

(b)

Difference between cover and steganographic audio signals: (a) Time Domain and (b) Frequency Domain.

B. Evaluation by typical attacks to the communication channel. Filtering, re-sampling, noise addition, echo addition and MPEG compression were used to evaluate the proposed algorithm. This evaluation was realized using “Audio StirMark 2.0 TM” and “Acoustica Audio Converter Pro 2005 TM software tools. When the re-sampling and compression test were realized the recovery percentages were minors who 40% and 80% respectively. In other applied tests the recovery percentage was greater than 95%. V. CONCLUSION DSSS technique was selected due its robustness features. The steganographic communication channel that use digital audio signals was evaluated strictly considering two aspects: functionality and robustness using different musical sorts in the MP3 files. When this technique is used, it is important to consider the attenuation that is applied to the pseudo noise additive, due the attenuation factor allows the imperceptibility of the subliminal information. The robustness tests were satisfactory except for the re-sampling and the compression of the cover signal due the recovery percentage was minor who 40% in both cases. ACKNOWLEDGEMENT This work was partially supported by the project SIP-IPN: 20082690. The scholarship granted to S. Herández-Garay is fully thanked to CONACYT REFERENCES [1.] W. Bender, D. Gruhl and A. Lu, “Techniques for data hiding (Journals style)”, IBM Systems Journal, vol. 35, Nos. 3 & 4, pp. 323-332, 1996. [2.] Litao Gang Akansu, "MP3 resistant oblivious steganography", in Proc IEEE International Conference on Acoustics, speech, and signal processing. (ICASSP´01). 2001. [3.] G. Kaliappan and W. Stanley and A. Scott and A. Darren, “Audio Steganography by Amplitude or Phase Modification,” in Proc. Security and Watermarking of Multimedia Contents V., 2003, Vol. 5020, pp 67-76. [4.] G. Kaliappan and S. Wenndt, “Audio Steganography for Covert Data Transmission by imperceptible Tone Insertion,” in Proc.the International Conference on Communication Systems and Application (CSA 2004), 2004, track 422-025. [5.] G. Kaliappan, “Audio steganography by cepstrum modification”, in Proc. of IASTED the International Conference on Acoustics, Speech, and Signal Processing (ICASSP´05), 2005, Vol. 5, pp 18-23. [6.] H. Matsuoka, “Spread Spectrum Audio Steganography using sub-band Phase Shifting”, in Proc. of International Conference on Intelligent Information Hiding and Multimedia (IIH-MSP´06), 2006, pp 3-6. [7.] G. Kaliappan “Audio steganography for embedding compressed speech”, in Proc of IASTED International Conference on Signal and Image Processing (SIP´02), 2002. USA.

978-1-4244-2284-5/08/$25.00 © 2008 IEEE

429