Efficient Watermarking of Audio Signals

Safaa Elgazar, Sami El-Dolil, Alaa M. Abbas, Fathi E. Abd El-Samie

Faculty of Electronic Engineering, Menoufia University, Menouf, Egypt
E-mail: fathi_sayed@yahoo.com

Abstract

This paper presents a comparative study between the watermark encryption

algorithms that are applied for audio signals. The chaotic Baker map and DES with its

four modes are used as watermark encryption algorithms. encrypted watermark is

embedded in audio signal using both SVD and block based SVD methods.

Experimental results show that the OFB mode is the least sensitive to attacks

compared with the chaotic Baker map.

1. Introduction

With the rapid development of the Internet, the transmission of digital multimedia

must achieve some requirements such as copyright protection and ownership

verification. Digital watermarking can use to achieve these requirements [1].

Watermarking is the process of embedding information in a cover signal (information

hiding) [2].

There are several methods used in the audio watermarking. One of these methods

is Singular Value Decomposition (SVD). It based on embedding the watermark in the

singular values (SVs) of the audio signal after transforming it into a 2-D format. After

watermark embedding, the audio signal is transformed again into a 1-D format. The

choice of SVs matrix is due to its stability against any variations which makes it

robust against attacks [3].

1
 To enhance the detectability of the watermark, the block based SVD algorithm is

used. The audio signal is divided into small segments and the watermark embedded in

the SVs of each segment separately. This method also increases robust against attacks

[3].

To achieve higher level of security and robustness against attacks, the watermark

is first encrypted and then embedded to the audio signal. There are several algorithms

to encrypt the watermark. One of these algorithms is the chaotic Baker map. it is one

of permutation based encryption algorithms, which are less sensitive to attacks [4,5].

The Data Encryption Standard (DES) is another encryption algorithm that can be used

for watermark encryption. It is a substitution – permutation algorithm. It characterized

by high level of security [6].

One of the most important requirement of the audio watermarking, that it should

not destroy the original signal but it should maintain the quality of the original signal.

So we use the Signal-to-Noise Ratio (SNR) to evaluate the quality of audio signal

after watermark embedding [7]. The main aim of the this paper is to present a

comparative study between watermarking encryption algorithms applied for audio

signals. The rest of the paper is organized as follows: Section 2, briefly explains the

chaotic encryption algorithm. Section 3, briefly explains the data encryption standard

(DES). Section 4, explains The Block-Based SVD Watermarking Approach. Section

5, presents the experimental results and Section 6, gives the concluding remarks

followed by the more relevant references.

2. Chaotic Baker Map Encryption

The Baker map is a chaotic map that generates a permuted version of a square

matrix. The baker map is an efficient tool to randomize a square matrix of data. Its

2
 operation is cut in half, and the two haves are stacked on one another. The baker

map, B, can be described with the following formulas [4,5]:

B( x, y )=(2 x , y /2 ) 0≤x< 1/ 2,
when (1)
B( x , y )=(2 x−1 , y /2+1/2) when 1/ 2≤x≤1 (2)

There are two versions of the chaotic baker map, generalized baker map and

discretized baker map. it depends on the secrete key k, it is a vector has k elements

such that the square matrix is divided into k vertical rectangles.

An example of permutation for a complete 8 pixel image is shown in Figure1. The

secret key is chosen to be (2-4-2) division, hence N= 8, n 1= 2, n2 = 4, and n3 = 2.

Figure (1)(a) explains the generalized Baker map and Figure (1)(b) explains the

discretized Baker map.

n n n
1 2 3

(a) Generalized Baker Map, N=8, n1 = 2, n2=4 and n3 =2.

(b) The Discretized Baker Map.

Fig. (1) The Baker map for an 8 pixel image by division (2, 4, 2).

3. Data Encryption Standard (DES)

3
 Data Encryption Standard (DES) is a block cipher that takes a plaintext (p) string

as input and creates a ciphertext (c) string of the same length. It uses a symmetric key

(k), which means that the same key is used to convert ciphertext back into plaintext.

The algorithm is designed to encryption and decryption blocks of data consisting of

64 bits under control of a 64-bit key, 8 bits of the key are used for error detection

which makes the effective DES key size 56 bits. Decryption process is the reverse of

the encryption process [8-10].

The encryption procedure is defined as:

C=E K ( P )
e
(3)
where Ke is the encryption key and E is the encryption function. Similarly, the
decryption procedure is defined as:
P=D K ( C )
d
(4)
Kd
where is the decryption key and D is the decryption function.
The encryption process can be briefly explained as follows : (1) Fractioning of the

data to be encrypted into 64-bit (8 octet) blocks. (2) Initial permutation IP of blocks.

(3) Breakdown of the blocks into two parts: left and right, named G and D. (4) A

complex key-dependent computation ,Permutation and substitution steps repeated 16

times (called rounds). (5) Re-joining of the left and right parts then inverse initial

permutation IP -1.

this round can be mathematically defined as:

Gn = Dn-1 (5)
Dn = Gn-1 f(Dn-1,Kn) (6)
Decryption process is the reverse of the encryption process. the round in decryption

can be mathematically defined as :

Dn-1 = Gn (7)
Gn-1 = Dn f(Gn,Kn) (8)

4
 Where f is the cipher function which operates on two blocks , one of 32 bits (left or

right part of the encrypted block) and one of 48 bits (key) and produce a block of 32

bits that subject to Exclusive OR with the other previous part.

There are four modes of operation for DES. these modes are ECB (Electronic

Codebook), CBC (Cipher block Chaining), CFB (Cipher feedback) and OFB (Output

feedback). Each mode has its advantages and disadvantages as explained clearly in

[11,12].

4. The Block Based SVD Watermarking Approach

The SVD audio watermarking scheme depends on embedding a single watermark

in the audio signal as a whole. If watermark is embedded several times as in the block

based SVD method, it is expected that it enhances the detectability of the watermark

and increase its robustness against attack. The block based SVD method based on

dividing the audio signal into small segments then embedding the watermark in the

SVs of each segment, separately, gives the chance that one or more of the watermarks

will be out of the attacks and a high correlation coefficient in the detection will be

obtained [1-3].

The watermark embedding process is illustrated in these steps:

1. The original audio signal is divided into non-overlapping segments and each

segment is transformed into a 2-D matrix.

2. The SVD is performed on the 2-D matrix of each segment (B i matrix) to obtain the

SVs (Si matrix), where i = 1, 2, 3, . . . , N, and N is the number of segments.

Bi=U i S i V Ti (9)

3. The encrypted watermark (W matrix) is added to the Si matrix of each segment.

5
 D i =S i + KW (10)

A small value of K of about 0.01 is required to keep the audio signal undistorted.

4. The SVD is performed on each Di matrix to obtain the SVs of each one (S wi

matrix).

Di =U wi S wi V Twi (11)

5. The Swi matrices are used to build the watermarked segments .

Bwi =U i Swi V Ti (12)

6. The watermarked segments are transformed into the 1-D format.

7. The watermarked segments are rearranged back into a 1-D audio signal.

Having Uwi, Vwi, Si, matrices and the possibly distorted audio signal, the steps

mentioned below can be followed to get the possibly corrupted watermark.

1. The possibly distorted watermarked signal is divided into small segments having

the same length used in the embedding process and these segments are

transformed into the 2-D format.

2. The SVD is performed on each possibly distorted watermarked segment (B* wi

matrix) to obtain the SVs of each one (S*wi matrix).

¿ ¿ ¿
Bwi =U i S wi V ¿i T (13)
3. The matrices that may contain the watermark are obtained using the U wi, Vwi,

S*wi, matrices.

D¿i =U wi S ¿wi V Twi (14)

4. The possibly corrupted encrypted watermarks (Wi* matrices) are extracted from the

Di matrices.

W ¿i =( D ¿i −Si ) / K
(15)
¿
5. The obtained matrix
W i matrices are decrypted.

6
 ¿
6. The correlation coefficient between each decrypted matrix W i and the original

watermark is estimated. If at least one of the coefficients is higher than a certain

threshold, the watermark is present.

5. Experimental Results

In this section, our experiments are carried out to test the chaotic baker map

encryption algorithm and data encryption standard (DES) with its four modes for the

addition of encrypted watermarks to the audio signals. We use both the SVD method

and the block based SVD method in our comparison. The CS image is used as a

watermark to be embedded on the Handel signal available in Matlab. In SVD method

we use the watermark image of dimensions 256  256, and hence an audio segment

of 65536 samples is used as the cover signal. And in the block based SVD method,

we use a small watermark of dimensions 16  16 in all segments of the audio signal,

The length of each segment is 256, which is the number of pixels in the small

watermark.

To measure the quality of encryption, we estimate the correlation coefficient

between the encrypted watermark and the original watermark images ce. The lower

the value of ce, the better the degree of the encryption. Also, to evaluate the quality of

watermarking, we estimate the correlation coefficient between the extracted

watermark and the original watermark images c r. The higher the value of C r, the

better the quality of watermarking.

Experiments are performed in the absence of attack and also in the presence of

different types of attacks. There are four types of attack on the watermarked audio

signal were studied . an additive white Gaussian noise (AWGN) attack, a low pass

7
filtering attack with a third order Butterworth filter, a cropping attack and a wavelet

compression attack.

Tables (1,3) give the values of ce and indicate that for the SVD method the CFB
mode is the best algorithm in the quality of encryption and for block based SVD
method the chaotic Baker map algorithm is the best.

Fig (2) shows the encrypted watermarks for SVD method. And table (2) gives

the values of c r. Fig (3) shows the encrypted watermarks for block based SVD

method. And table (4) gives the values of cmax. we notice that, In the absence of

attacks, all encryption algorithms achieve high correlation coefficient where cr =1

and perfect extraction of watermark in both SVD and block based SVD methods.

In the presence of attacks, for the SVD method, the OFB mode is the lest affected

to all types of attacks followed by the chaotic baker map. And for the block based

SVD method, also the OFB is the least affected to noise and wavelet compression

attacks followed by chaotic Baker map. The chaotic Baker map is most of them

affected by filtering attack unlike CFB. We notice that the watermark can be perfect

extracted in the presence of cropping attack in the block based method. This due to

the repeat of embedding the watermark in each segment of the audio signal but SVD

method embed only one watermark embedded in the whole audio signal.

In our experiments Signal-to-Noise Ratio is used to evaluate the quality of the

watermarked audio signal. In the absence of attacks, we notice that both the SVD and

block based SVD method audio watermarking doesn't degrade the quality of the

watermarked audio signal. But presence of attacks affect the watermarked audio

signal and reduce the SNR values and this is indicated in tables (5,6).

Fig (4) show the variation of the correlation coefficient Cr with the SNR in the

presence of white noise attack, and it indicates that OFB mode gives higher values of

8
Cr. Fig (5) show the variation of the correlation coefficient Cr with the filtering BW in

the presence of the filtering attack, and it indicates the OFB mode gives higher values

of Cr for SVD method and the CFB mode gives higher values for block based SVD

method. Fig (6) show the variation of the correlation coefficient Cr with the

compression threshold in the presence of wavelet compression attack, and it indicates

that OFB mode gives higher values of Cr.

9
 CBC Chaotic Original watermark

OFB ECB CFB

Fig. (2) original watermark and encrypted watermarks using chaotic Baker map
and DES encryption modes using the SVD method.

Table (1) Correlation coefficient (ce) between the original watermark and the
encrypted watermark.
Correlation
Chaotic CBC CFB ECB OFB
coefficient
ce 0.0181 0.0027 - 0.0041 - 0.2852 0.0013

Table (2) Correlation coefficient (cr) between the decrypted extracted watermark
and the original watermark using the SVD method with the chaotic Baker map
and the DES encryption modes.
OFB ECB CFB CBC Chaotic Encryption
algorithm/attack
type
1 1 1 1 1 No attacks
0.6212 - 0.0108 0.0037 0.0036 0.0568 Noise
0.0167 - 0.0981 - 0.007 - 0.0034 0.0048 Filtering
0.3278 - 0.0040 - 0.0042 - 0.0035 0.0307 Cropping
0.5856 0.0058 0.0029 -7.32e -004 0.1251 Wavelet
compression

10
 CBC Chaotic Original watermark

OFB ECB CFB

Fig. (3) original watermark and encrypted watermarks using chaotic Baker map
and DES encryption modes using the block-based SVD method.

Table (3) Correlation coefficient (ce) between the original watermark and the

encrypted watermark using the block-based SVD method.

Correlation
chaotic CBC CFB ECB OFB
coefficient
ce - 0.0572 - 0.0351 2.87e-004 - 0.0206 0.0021

Table (4) Maximum correlation coefficient (cmax) between the decrypted

extracted watermark which achieves maximum correlation coefficient and the
original watermark using the block based SVD method with the chaotic Baker
map and the DES encryption modes.
OFB ECB CFB CBC Chaotic Encryption
algorithm/attack
type
1 1 1 1 1 No attacks
0.6794 0.1615 0.1642 0.1718 0.1736 Noise
0.1273 0.1613 0.2110 0.1940 - 0.026 Filtering
1 1 1 1 1 Cropping
0.7655 0.2077 0.1986 0.1666 0.2239 Wavelet
Compression

11
Table (5) SNR (dB) between original audio signal and watermarked audio signal
using the SVD method.

OFB ECB CFB CBC Chaotic Encryption

algorithm/attack
type
52.6908 54.64 52.0839 52.7526 48.0238 No attacks
- 9.9976 - 10.0041 - 10.0424 - 10.0017 - 10.0467 Noise
1.5582 1.5579 1.5583 1.5582 1.5599 Filtering
3.0433 3.0434 3.0433 3.0433 3.0433 Cropping
Wavelet
9.1558 9.1559 9.1558 9.1558 9.1556
Compression

Table (6) SNR (dB) between original audio signal and watermarked audio signal
using the block based SVD method.

Encryption
OFB ECB CFB CBC Chaotic algorithm/attac
k type
41.1200 43.4939 44.1727 43.6775 37.4362 No attacks
- 10.0055 - 10.0456 - 10.0238 - 10.0592 - 10.0978 Noise
1.5635 1.5633 1.5634 1.5626 1.5693 Filtering
3.0430 3.0432 3.0432 3.0432 3.0426 Cropping
Wavelet
9.1544 9.1551 9.1550 9.155 9.1574
Compression

12
 (a) SVD method

(b) Block based SVD method

Fig. (4) Variation of the correlation coefficient cr with the SNR in the presence of
white noise attack.

13
 (a) SVD method

(c) Block based SVD method

Fig. (5) Variation of the correlation coefficient cr with the filtering BW in the
presence of the filtering attack.

14
 (a) SVD method

(b) Block based SVD method

Fig. (6) Variation of the correlation coefficient cr with the compression threshold
in the presence of wavelet compression attack.

15
6. Conclusions

This paper presented a comparative study of two encryption algorithms; the chaotic
Baker map encryption algorithm and the DES encryption algorithm with its four
modes (CBC , CFB, ECB , OFB). Encrypted watermark embedded in both SVD
method and block based SVD method. Experiments shows that for the SVD method
the CFB mode is the best algorithm in the quality of encryption and for block based
SVD method the chaotic Baker map algorithm is the best. The OFB mode is the least
sensitive to attacks using the two methods of watermarking compared with the chaotic
Baker map.

References

[1] Chin-Chen Chang, Yih-Shin Hu and Chia-Chen Lin, "A Digital Watermarking

Scheme Based on Singular Value Decomposition", Springer Berlin Heidelberg,

2007.

[2] F. E. Abd El-samie, “An efficient singular value decomposition algorithm for

digital audio watermarking, “ International Journal of Speech Technology, Vol.

12, No. 1, pp. 27–45, 2009, Springer.

[3] W. Al-Nuaimy, M. A.M. El-Bendary, A. Shafik, F. Shawki, A.E. Abou-El-azm,

N.A. El-Fishawy, S. M. Elhalafawy, S. M. Diab, B. M. Sallam, F. E. Abd El-

Samie, and H.B. Kazemian, "An SVD Audio Watermarking Approach Using

Chaotic Encrypted Images," Digital Signal Processing, Vol. 21, No. 6, pp. 764-

779, 2011, Elsevier

[4] J. Fridrich, “Image Encryption Based On Chaotic Maps" Proceedings of the IEEE

International Conference on Systems, Man, and Cybernetics, vol.2, pp. 1105-

1110, 1997.

[5] J. Fridrich,” Symmetric Ciphers Based on Two-dimensional Chaotic Maps” Int.J.

Bifurcation and Claos, vol. 8, No.6, pp. 1259-1284, 1998.

16
[6] Dr Jean-Yves Chouinard, "Design of Secure Computer System

CSI4138/CEG4394 . Notes on the Data Encryption Standard (DES)", uottawa.ca,

2002.

[7] K. Kondo, " Subjective Quality Measurement of Speech, Signals and

Communication Technology", Springer-Verlag Berlin Heidelberg, 2012.

[8] F. PUB. Data Encryption Standard (DES).National Bureau of Standards, US

Department of Commerce, 1977.

[9] http://en.kioskea.net/contents/134-introduction-to-encryption-with-des,

(access date 29/7/213).

[10] Ghanshyam Gaur1, R. S. Meena, "Secure Transmission of Biometric Scan

Images Using Data Encryption Standards(DES) Algorithm", Scientific &

Academic Publishing, 2012.

[11] http://en.wikipedia.org/wiki/Block_cipher_mode_of_operation,

(access date 29/7/2013).

[12] Ibrahim Fathy El-Ashry, "Literature Survey on Image Security chapter 2 of M.

Sc. Thesis Submitted for The Degree of M. Sc. of Communications Engineering",

Faculty of Electronic Engineering, 2010.

17