# Robust reversible watermarking based on normalized correlation quantization index modulation combined with Cdma techniques.

INTRODUCTION

In the current trends of the world, due to the recent advancement, most of the individuals prefer to use the internet as the primary medium to transfer data from one region to another region across the whole world. The data transmission is made very simple, fast and accurate using the net facility. However, security is the main issue while sending data over the internet. The private content can be hacked in various ways. Therefore it becomes required to take data security into consideration. Data security means protection or safeguard of data from unauthorized users or hackers and providing high security to prevent data alteration. In order to overcome the problem of the security attacks in data transfers over the internet, digital image watermarking technique is used. It is one of the prominent methods to fulfill the gap between copyright issues and digital distribution of data content. In addition to copyright protection, watermarking has many other applications, including authentication, access control, broadcast technology, communication lines. One of the most important techniques proposed is Quantization Index Modulation (QIM) [12], where watermark embedding is obtained by quantizing the host or input signal with a quantizer level chosen among a full set of quantizers that are each associated with a different message. The basic implementation method of QIM, is known as Dither Modulation (DM), adopts a large set of scalar and uniform quantizers [12]. Later, the distortion compensation method was combined with the QIM method, resulting in the Distortion Compensated QIM (DC-QIM) [12]. DC-QIM follows Costa's procedures and closes the gap to capacity. The Scalar Costa Scheme (SCS) proposed in [6] can be regarded as a special example of DC-QIM. The theoretical presentation of QIM methods has been extensively investigated [6], [7]. In these basic type algorithms, the amplitude values of one single pixel or of a set of pixels are straightforwardly quantized. Digital watermarking method can be observed as a communication issue with side information [3]. The side information is knowledge of the host signal content, which is obtainable to the watermark embedder. The channel distortion level creates an unknown noise source. For this issue, an important result was given by Costa [4] that the capacity of a channel with an additive Gaussian noise and some of the extra noise relevant known data to the encoder is the similar as that of the channel in that this extra noise is absent state

The Logarithmic QIM (LQIM), performing a logarithmic transform method on the input signal before quantization [8, 9]. Due to the use of logarithmic transformation, LQIM method obtains non-uniform quantization step sizes for watermark embedding step, which makes it have perceptual benefits. Some gain invariant features have been utilized in quantization-based watermarking method with a particular significance on Valumetric Scaling Attack (VSA) [10]. Typically, the signal for quantization was constructed by the method of dividing the host signal value by its arithmetic mean in [11]. The angle feature formed by the host signal vector was used for quantization in Angle QIM (AQIM) [12]. The idea of AQIM was also applied to improve the performance of STDM in [13]. Normalized Correlation based Quantization Index Modulation (NCQIM) is the extension of QIM [16, 17]. Normalized Correlation based Quantization Index Modulation (NCQIM) is the extension of QIM [16, 17].

NCQM [1] has been also extensively tested on various images and compared with the state-of-the-art watermarking methods. Experimental results demonstrate that the method obtains the improved imperceptibility and a relatively high embedding capacity in a very noisy environment. Moreover, it yields superior robustness against a wide range of common image processing operations, such as VSA, Gaussian filtering, AWGN and Gamma correction, etc. The limitation of NCQM is that it presents the significant robustness advantage only for the host vector dimension being large enough.

In this paper, we refer to as Normalized Correlation based Quantization Index Modulation (NCQIM) Combined with Least Significant Bit (LSB) and Code Division Multiple Access (CDMA) techniques that is, in general, many specific situations optimal. Where the feature signals for modulation is the Normalized Correlation formed by the host signal and a random signal. To perform the information level modulation, the structured codebooks are designed using uniform quantizes. The watermarked signal is created by the embedding function in order to form the modulated NC with the random signal. The LSB and CDMA method is used before embedding the watermarked message. The watermark imperceptibility and robustness of this method is achieved using the parameters Peak Signal to Noise Ratio (PSNR) and Bit Error Rate (BER) respectively.

II. Related Works:

In digital watermarking, the side information is knowledge of the host signal, which is available to the watermark embedded. I.J. Cox, M.L. Miller, and A. McKellips suggested [1] that perceptual modeling had significant utility for watermarking. If a watermark is to be robust, i.e. survive, common signal processing that is routinely performed on the same content, then the watermark signal must be employed in the perceptually significant components of the content. They suggested that the watermark signal should be modified based on the perceptual properties of the content. A perceptual model is often used to increase the power of the signal while maintaining fidelity by amplifying the signal wherever the model's sensitivity is low. Brian Chen and Gregory W. Wornell suggested [4] that a new class of embedding methods, termed QIM and Distortion Compensated QIM (DCQIM) Using deterministic models to evaluate digital watermarking methods, they show that QIM method is provably good against arbitrary bounded and fully informed attacks, which arise in several copyright secure applications, it achieves provably better rate distortion robustness tradeoffs than currently popular SS and Low Bit Modulation (LBM) methods. This method has purely additive function.

J.J. Eggers, R. Bauml, R. Tzschoppe and B. Girod, [25] give a complete performance analysis of the Scalar Costa Scheme (SCS), which is a suboptimal techniques using scalar embed and reception functions. Information theoretic bounds and simulation results with state of the art type coding techniques are compared. Further, reception after amplitude scaling attacks and the invertibility of SCS embedding are investigated. Thus, information embedding might be a more correct term than digital watermarking since the robustness requirement is weakened to robustness against AWGN. F. Bartolini, M. Barni and A. Piva shows that, the performance of STDM watermarking [5] in the presence of two important classes of non-additive attacks, such as the gain attack with noise addition, and the quantization attack are evaluated. The analysis is developed under the assumption that the host features are IID Gaussian random values, and a minimum distance criterion is used to decode the hidden information.

F. Ourique, V. Licks, R. Jordan and F. Perez-Gonzalez shows that, AQIM [10] works by quantizing the angle formed by the host signal vector with the origin of a hyper spherical coordinate system. Hence, AQIM's invariance to amplitude scaling can be shown by construction. In addition to amplitude scaling robustness, AQIM's performance to AWGN attacks was assessed experimentally by means of Monte Carlo experts simulations. Experimental value results are presented for the bit error rate performance of AQIM under AWGN attacks. F. Balado suggested that, a new geometric interpretation of additive and multiplicative SS watermarking [22] with repetition coding and decoding is used. The interpretation gives an intuitive rational on why the multiplicative scheme achieves better in additive independent attacks, and it is also used to produce an original quantitative performance analysis.

V.H. Mankar, T.S. Das and S.K. Sarkar solved the problem based on Angle Quantization Index Modulation (AQIM) within STDM framework [11]. AQIM method embeds the information by quantizing the angle formed by the host-signal vector with respect to the origin of a hyper spherical coordinate system as opposed to quantizing the amplitude of pixel values. The AQIM method, which keeps the amplitude value of pixels unchanged, is completely invariant to scaling distortion theoretically. A new method for LQIM is proposed by N.K. Kalantari and S.M. Ahadi [5], in this a logarithmic function is first applied to the host or input signal. Then the transformed signal is quantized using uniform quantization as conventional QIM to embed watermark data within that. Finally using inverse transform the watermarked signal is achieved. The watermark extraction is done using minimum distance decoder. The optimum parameter for data embedding with minimum quantized distortion is derived. It is used for only vector quantization.

III. Proposed Method:

In this paper, the NC-QIM, LSB and CDMA techniques are combined for embedding and extracting the watermarked message. This method is used to increase the robustness of the watermarked image. Most schemes require that the watermark be recovered even if the watermarked image is distorted and this distortion may be accidental or deliberate. So the watermarked image should be robust against any attacks. The host signal or input signal is transformed into the feature signal. In this transformation, the normalized correlation is calculated between the host signal and the random signal. Then the modulation is performed on the feature signal and watermark message. The LSB method is applied in this modulated signal. Then CDMA method is applied. Then the embedding function is performed to produce the watermarked signal. This watermarked signal goes through the channel. At decoder side, the received signal is transformed into feature signal by using NCQIM technique which is done at embedded side and LSB method is used. Then the watermarked message is extracted from feature signal.

3.1 Normalized Correlation Based Quantization Index Modulation:

In the design methodology of the feature modulation watermarking method, a serious issue is to generate the feature signal from the host signal. For this purpose, the features should be selected by taking into account the properties of the utilized modulation technique. The feature signals are obtained by using PRNG seeded with the secret key. The host feature signal ([f.sub.x]) by computing the normalized correlation of the host vector (x) and a random vector (u) is,

[f.sub.x] = [x.sup.T]u/[parallel]x[parallel] [parallel]u[parallel] (1)

PRNG is an algorithm for creating a sequence of numbers whose properties almost equal to the properties of sequences of random numbers. It is completely calculated by a small set of initial values called the PRNG's seed. Pseudo random numbers are created by providing a random seed value to PRNG. Normalization is a process that changes the range of pixel intensity values. In watermarking, the normalized correlation is calculated between the host signal and random signal. The random signals are made by a PRNG seeded with the secret key, and also used for the generation of host signals and watermarked signals. This increases the probability of correct detection.

3.2 quantization index modulation:

The QIM used to embedding information by modulating a set of indices with the embedded information and then quantizing the input value with the related sequence of quantizes. In this method the amplitude of single pixel (or) a vector of pixels is quantized using quantizes. Given the quantization step A and a dither value d [epsilon] [[DELTA]/4, [DELTA]/4], the sets [[conjunction].sub.0] and [[conjunction].sub.1] are respectively represented by,

[[conjunction].sub.0] = {k[DELTA] + d/k [epsilon] z} [intersection] [-1, 1] (2)

[[conjunction].sub.0] = (k[DELTA] + d + 1/2[DELTA]\k [epsilon] z} [intersection] [-1, 1] (3)

This is used for normalized correlation. The normalized correlation [f.sub.x] is then modulated with the hidden symbol m by selecting a code word fm from the corresponding codebook [[conjunction].sub.m] to replace [f.sub.x.]. This can be implemented by a quantization operation.

3.2.1 Lsb Based Watermarking:

In this technique, the watermarking process is done in two methods: Watermarking Embedding techniques and Watermark Recovery techniques. In watermarking embedding the images are process in pixels and generate the watermark to cover original signal. In the embedding process, the size of the original image is obtained. Then the maximum message size to be embedded is determined based on the original image. Next the size of the key image is calculated. Then the watermark image is reshaped into vector. After this, the size of the watermark image is smaller than that of cover image. The watermark image is embedded into the cover image using LSB. Watermark extraction is done by the reverse process.

3.2.1 Cdma Based Watermarking:

In this, we propose a CDMA based multiple-user sharing-channel watermarking technique which is useful for applications in multiple-user sharing copyright protection. Although there are many watermarking algorithms related to CDMA, these single watermarking algorithms only utilize the spread spectrum characteristic of CDMA to enhance the robustness of the algorithms and do not make full use of its potential peculiarity of supporting multiple-user sharing channel.

[FIGURE 1 OMITTED]

To detect the watermark, each seed is used to generate its PN sequence, which is then correlated with the complete image. If the correlation level is high, that bit in the watermark is set to "1", otherwise a "0". The same method is then repeated for all the values of the watermark. CDMA improves on the robustness of the watermark significantly, but requires several orders more of calculation. In this, a novel approach based on the unconventional Random Encoding SS method is proposed for recovering the spreading sequence of watermark signal without any information from the transmitter. It is contribute a higher secure feature by using of the time varying random encoded spread spectrum.

3.3 Watermark Embedding And Extraction Function:

The embedding function Em (.) is designed to produce a watermarked signal. In this model, the information modulation is not directly done on the input signal itself, but based on a feature signal transformed method from the host signal, which is quite different from the existing watermarking methods. The performance improvement can be brought about if an appropriate feature is utilized. The watermark signal is embedded in the original image after applying LSB technique. The watermarked signal goes through a particular channel, where it might be exposed to several common signal processing manipulations. The Gaussian noise is added in the channel to reduce the losses. This Gaussian noise is added in the channel to produce distorted watermarked signal.

The distorted watermarked signal can be written as,

z = y + n (4)

Where, z - Distorted watermarked signal, Y - Watermarked signal,

3.4 Watermark Extraction:

At the decoder side, the received signal z is first transformed into the feature signal [f.sub.z] as done at the embedded side. Then, a message [??] is extracted from [f.sub.z] according to the correspondence between the watermark content and the used the same feature level, which is established during the information modulation process and a message is extracted from [f.sub.z].

4 Parameter Analysis:

4.2 Watermark Imperceptibility:

Imperceptibility means that the watermarked content is perceptually indistinguishable from the original image. This property is related to the robustness of the watermark and hence an optimal balance between imperceptibility and robustness must be achieved by the watermarking scheme. Imperceptibility can be measured using Peak Signal to Noise Ratio (PSNR) between host and watermarked image and Root Mean Squared Error (RMSE). The PSNR value is calculated by using Mean Squared Error (MSE).

4.2.1 Peak Signal to Noise Ratio:

The visual quality of the embedded images can also be measured using the PSNR. It is the most easily defined by the Mean Squared Error (MSE). The PSNR is defined as

PSNR = 10[log.sub.10] ([MAX.sup.2]/MSE) in decibels (5)

Where, MAX--Maximum pixel value of the frame, MSE--Mean Squared Error

The MSE measures the statistical difference in the pixel values between the original and the reconstructed image. The MSE is defined as,

MSE(x,y) = 1/MN [[summation].sup.M.sub.y=1] [[summation].sup.N.sub.x=1] ][I(x,y) [I.sup.I] (x, y)].sup.2] (6)

Where,

X--Original image, Y--Watermarked image, M, N--Size of the original image.

The RMSE is a quadratic scoring rule which measures the average magnitude of the error. It is calculated by using the formula,

RMSE = sqrt(MSE) (7)

4.3 Robustness:

The robustness can be achieved by using the parameter BER. Bit Error Rate is determined by Inverse of PSNR values.

BER = 1/PSNR (8)

The more is BER lesser will be the quality of the Watermarking technique. So BER is inversely proportional to PSNR. More the value of BER lesser will be PSNR value.

5 Result Analysis:

In this paper the simulation process is implemented in MATLAB using different types of host images and watermark images. For analysis purpose gray scaled image is used. The opted host (original) and watermark images are as follows: The performance evaluation can be analyzed by using the Peak Signal to Noise Ratio (PSNR). The following figures represent the extracted watermark images with several attacks of the proposed scheme. The different levels of values of PSNR & MSE for different images are listed in Table 2 and 3.

[TABLE 1 OMITTED]

Also for different images the values of RMSE and Bit Error Rate also changes. The different values of Root Mean Square Error and Bit Error Rate for different images have been listed in Table 4 and 5.

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

[FIGURE 5 OMITTED]

Conclusion:

In this paper, a novel quantization based watermarking method, called NCQIM based LSB combined with CDMA technique, has been developed. In the method, the NC between the host signal and a random signal was modulated with the hidden message using the structured codebooks, and then the watermarked signal was formed to provide the modulated NC by minimizing the embedding distortion. The induced watermark signal by NCQIM is a linear combination of the host signal and the random signal, which is perceptually better than random watermark signals independent of the host signal. Moreover, we theoretically evaluated the performance of the proposed method. According to our analyses, the embedding distortion of NCQIM is a monotonically decreasing function of the quantization step and proportional to the host signal variance. NCQIM has been also extensively tested on various images and compared with the state-of-the-art watermarking methods. Experimental results demonstrate that our method obtains the improved imperceptibility and a relatively high embedding capacity in a very noisy environment. Moreover, it yields superior robustness against a wide range of common image processing operations. This proposed method further can be extended with various embedding algorithm to enhance the robustness level.

REFERENCES

[1.] Xinshan Zhu, Jie Ding, and Honghui Dong, 2014. "Normalized Correlation-Based Quantization Modulation For Robust Watermarking" IEEE Transactions On Multimedia, 16: 7.

[2.] Zhuand, X., J. Ding, 2012. "A novel quantization watermarking scheme using random normalized correlation modulation," Chin.J.Comput., 35(9): 1959-1970.

[3.] Zhu, X. and S. Peng, 2012. "A novel quantization watermarking scheme by modulating the normalized correlation," in Proc. Int. Conf. Acoust., Speech Signal Process., pp: 1765-1768.

[4.] Guerrini, F., M. Okuda, N. Adami and R. Leonardi, 2011. "High dynamic range image watermarking robust against tone-mapping operators," IEEE Trans. Inf. Forensics Security, 6(2): 283-295.

[5.] Kalantari, N.K. and S.M. Ahadi, 2010. "A logarithmic quantization index modulation for perceptually better data hiding," IEEE Trans. Image Process., 19(6): 1504-1517.

[6.] Kalantari, N.K., S.M. Ahadi and M. Vafadust, 2010. "A robust image watermarking in the ridge let domain using universally optimum decoder," IEEE Trans. Circuits Syst. Video Technol., 20(3): 396-406.

[7.] Mankar, V.H., T.S. Das, and S.K. Sarkar, 2009. "An angle QIM watermarking in STDM framework robust against amplitude scaling distortions," Contemporary Comput., Commun. Comput. Inf. Sci., 40(8): 400-410.

[8.] Cox, I.J., M. Miller, J.A. Bloom, J. Fridrich and T. Kalker, 2008. Digital Watermarking and Steganography, 2nd ed. San Francisco, CA, USA: Morgan Kaufmann.

[9.] Comesana, P. and F. Perez-Gonzalez, 2007. "On a watermarking scheme in the logarithmic domain and its perceptual advantages," in Proc. IEEE Int. Conf. Image Process., San Antonio, TX, USA, pp: 145-148.

[10.] Li, Q. and I.J. Cox, 2007. "Using perceptual models to improve fidelity and provide resistance to valumetric scaling for quantization index modulation watermarking, " IEEE Trans. Inf. Forensics Security, 2(2): 127-139.

[11.] Lan, T.H., A.H. Tewfik, 2006. "A Novel High-Capacity Data- Embedding System," IEEE Trans. On Image Processing, 15(8): 2431-2440.

[12.] Perez-Gonzalez, F., C. Mosquera, M. Barni and A. Abrardo, 2005. "Rational dither modulation: A high-rate data-hiding method invariant to gain attacks," IEEE Trans. Signal Process., 53(10): 3960-3975.

[13.] Ourique, F., V. Licks, R.. Jordan and F. Perez-Gonzalez, 2005. "Angle qim: A novel watermark embedding scheme robust against amplitude scaling distortions," in Proc. Int. Conf. Acoust., Speech Signal Process, pp: 797-800.

[14.] Moulin, P. and R. Kotter, 2005. "Data-Hiding Codes," Proceedings IEEE, 93(12): 2083-2127.

[15.] Celik, M., G. Sharma, A. Tekalp and E. Saber, 2005. "Lossless generalized-LSB data embedding," IEEE Trans. Image Process., 14(2): 253-26.

[16.] Balado, F., 2005. "New geometric analysis of spread-spectrum data hiding with repetition coding, with implications for side-informed schemes," in Proc.Int.Workshop Digital Watermark., pp: 336-350.

[17.] Bartolini, F., M. Barni and A. Piva, 2004. "Performance analysis of ST-DM watermarking in presence of non-additive attacks," IEEE Trans. Signal Process, 52(10): 2965-2974.

[18.] Wu, M. and B. Liu, 2003. "Data hiding in image and video: Part I-fundamental issues and solutions," IEEE Trans. Image Process., 12(6): 685-695.

[19.] Perez-Gonzalez, F., F. Balado and J.R.H. Martin, 2003. "Performance analysis of existing and new methods for data hiding with known-host information in additive channels," IEEE Trans.Signal Process., 51(4): 960-980.

[20.] Eggers, J.J., R. Bauml, R. Tzschoppe and B. Girod, 2003. "Scalar Costa scheme for information embedding," IEEE Trans. Signal Process, 51(4): 1003-1019.

[21.] Fridrich, J., M. Goljan and R. Du, 2002. "Lossless data embedding-new paradigm in digital watermarking," EURASIP J. Appl. Signal Process., 1: 185-196.

[22.] Chen, B. and G.W. Wornell, 2001. "Quantization index modulation: A class of provably good methods for digital watermarking and information embedding," IEEE Trans. Inf. Theory, 47(4): 1423-1443.

[23.] Chen, B. and G.W. Wornell, 2001. "Implementations of quantization index modulation methods for digital watermarking and information embedding of multimedia," J. VLSI Signal Processing Syst. Signal, Image, and Video, Technology (Special Issue on Multimedia Signal Processing), 27: 7-33.

[24.] Digital Image Processing using Matlab By, Gonzalez

[25.] Cox, I.J., M.L. Miller and A. McKellips, 1999. "Watermarking as communications with side information," Proc. IEEE, 87(7): 1127-1141.

[26.] Deller, J.R., Jr., J.G. Proakis and J.H.L. Hansen, 1993. Discrete Time Processing of Speech Signals, New York: Macmillan.

(1) M. Subbulakshmi and (2) M. Vijayaraj

(1) Department of ECE, National College of Engineering, Maruthakulam-627151, Tamilnadu, India.

(2) Department of ECE, Government College of Engineering, Tirunelveli-627007, Tamilnadu, India.

Received 15 May 2016; Accepted 7 July 2016; Available 22 July 2016

Address For Correspondence:

M. Subbulakshmi, Department of ECE, National College of Engineering, Maruthakulam-627151, Tamilnadu, India.

E-mail: subbu_june23@yahoo.co.in

In the current trends of the world, due to the recent advancement, most of the individuals prefer to use the internet as the primary medium to transfer data from one region to another region across the whole world. The data transmission is made very simple, fast and accurate using the net facility. However, security is the main issue while sending data over the internet. The private content can be hacked in various ways. Therefore it becomes required to take data security into consideration. Data security means protection or safeguard of data from unauthorized users or hackers and providing high security to prevent data alteration. In order to overcome the problem of the security attacks in data transfers over the internet, digital image watermarking technique is used. It is one of the prominent methods to fulfill the gap between copyright issues and digital distribution of data content. In addition to copyright protection, watermarking has many other applications, including authentication, access control, broadcast technology, communication lines. One of the most important techniques proposed is Quantization Index Modulation (QIM) [12], where watermark embedding is obtained by quantizing the host or input signal with a quantizer level chosen among a full set of quantizers that are each associated with a different message. The basic implementation method of QIM, is known as Dither Modulation (DM), adopts a large set of scalar and uniform quantizers [12]. Later, the distortion compensation method was combined with the QIM method, resulting in the Distortion Compensated QIM (DC-QIM) [12]. DC-QIM follows Costa's procedures and closes the gap to capacity. The Scalar Costa Scheme (SCS) proposed in [6] can be regarded as a special example of DC-QIM. The theoretical presentation of QIM methods has been extensively investigated [6], [7]. In these basic type algorithms, the amplitude values of one single pixel or of a set of pixels are straightforwardly quantized. Digital watermarking method can be observed as a communication issue with side information [3]. The side information is knowledge of the host signal content, which is obtainable to the watermark embedder. The channel distortion level creates an unknown noise source. For this issue, an important result was given by Costa [4] that the capacity of a channel with an additive Gaussian noise and some of the extra noise relevant known data to the encoder is the similar as that of the channel in that this extra noise is absent state

The Logarithmic QIM (LQIM), performing a logarithmic transform method on the input signal before quantization [8, 9]. Due to the use of logarithmic transformation, LQIM method obtains non-uniform quantization step sizes for watermark embedding step, which makes it have perceptual benefits. Some gain invariant features have been utilized in quantization-based watermarking method with a particular significance on Valumetric Scaling Attack (VSA) [10]. Typically, the signal for quantization was constructed by the method of dividing the host signal value by its arithmetic mean in [11]. The angle feature formed by the host signal vector was used for quantization in Angle QIM (AQIM) [12]. The idea of AQIM was also applied to improve the performance of STDM in [13]. Normalized Correlation based Quantization Index Modulation (NCQIM) is the extension of QIM [16, 17]. Normalized Correlation based Quantization Index Modulation (NCQIM) is the extension of QIM [16, 17].

NCQM [1] has been also extensively tested on various images and compared with the state-of-the-art watermarking methods. Experimental results demonstrate that the method obtains the improved imperceptibility and a relatively high embedding capacity in a very noisy environment. Moreover, it yields superior robustness against a wide range of common image processing operations, such as VSA, Gaussian filtering, AWGN and Gamma correction, etc. The limitation of NCQM is that it presents the significant robustness advantage only for the host vector dimension being large enough.

In this paper, we refer to as Normalized Correlation based Quantization Index Modulation (NCQIM) Combined with Least Significant Bit (LSB) and Code Division Multiple Access (CDMA) techniques that is, in general, many specific situations optimal. Where the feature signals for modulation is the Normalized Correlation formed by the host signal and a random signal. To perform the information level modulation, the structured codebooks are designed using uniform quantizes. The watermarked signal is created by the embedding function in order to form the modulated NC with the random signal. The LSB and CDMA method is used before embedding the watermarked message. The watermark imperceptibility and robustness of this method is achieved using the parameters Peak Signal to Noise Ratio (PSNR) and Bit Error Rate (BER) respectively.

II. Related Works:

In digital watermarking, the side information is knowledge of the host signal, which is available to the watermark embedded. I.J. Cox, M.L. Miller, and A. McKellips suggested [1] that perceptual modeling had significant utility for watermarking. If a watermark is to be robust, i.e. survive, common signal processing that is routinely performed on the same content, then the watermark signal must be employed in the perceptually significant components of the content. They suggested that the watermark signal should be modified based on the perceptual properties of the content. A perceptual model is often used to increase the power of the signal while maintaining fidelity by amplifying the signal wherever the model's sensitivity is low. Brian Chen and Gregory W. Wornell suggested [4] that a new class of embedding methods, termed QIM and Distortion Compensated QIM (DCQIM) Using deterministic models to evaluate digital watermarking methods, they show that QIM method is provably good against arbitrary bounded and fully informed attacks, which arise in several copyright secure applications, it achieves provably better rate distortion robustness tradeoffs than currently popular SS and Low Bit Modulation (LBM) methods. This method has purely additive function.

J.J. Eggers, R. Bauml, R. Tzschoppe and B. Girod, [25] give a complete performance analysis of the Scalar Costa Scheme (SCS), which is a suboptimal techniques using scalar embed and reception functions. Information theoretic bounds and simulation results with state of the art type coding techniques are compared. Further, reception after amplitude scaling attacks and the invertibility of SCS embedding are investigated. Thus, information embedding might be a more correct term than digital watermarking since the robustness requirement is weakened to robustness against AWGN. F. Bartolini, M. Barni and A. Piva shows that, the performance of STDM watermarking [5] in the presence of two important classes of non-additive attacks, such as the gain attack with noise addition, and the quantization attack are evaluated. The analysis is developed under the assumption that the host features are IID Gaussian random values, and a minimum distance criterion is used to decode the hidden information.

F. Ourique, V. Licks, R. Jordan and F. Perez-Gonzalez shows that, AQIM [10] works by quantizing the angle formed by the host signal vector with the origin of a hyper spherical coordinate system. Hence, AQIM's invariance to amplitude scaling can be shown by construction. In addition to amplitude scaling robustness, AQIM's performance to AWGN attacks was assessed experimentally by means of Monte Carlo experts simulations. Experimental value results are presented for the bit error rate performance of AQIM under AWGN attacks. F. Balado suggested that, a new geometric interpretation of additive and multiplicative SS watermarking [22] with repetition coding and decoding is used. The interpretation gives an intuitive rational on why the multiplicative scheme achieves better in additive independent attacks, and it is also used to produce an original quantitative performance analysis.

V.H. Mankar, T.S. Das and S.K. Sarkar solved the problem based on Angle Quantization Index Modulation (AQIM) within STDM framework [11]. AQIM method embeds the information by quantizing the angle formed by the host-signal vector with respect to the origin of a hyper spherical coordinate system as opposed to quantizing the amplitude of pixel values. The AQIM method, which keeps the amplitude value of pixels unchanged, is completely invariant to scaling distortion theoretically. A new method for LQIM is proposed by N.K. Kalantari and S.M. Ahadi [5], in this a logarithmic function is first applied to the host or input signal. Then the transformed signal is quantized using uniform quantization as conventional QIM to embed watermark data within that. Finally using inverse transform the watermarked signal is achieved. The watermark extraction is done using minimum distance decoder. The optimum parameter for data embedding with minimum quantized distortion is derived. It is used for only vector quantization.

III. Proposed Method:

In this paper, the NC-QIM, LSB and CDMA techniques are combined for embedding and extracting the watermarked message. This method is used to increase the robustness of the watermarked image. Most schemes require that the watermark be recovered even if the watermarked image is distorted and this distortion may be accidental or deliberate. So the watermarked image should be robust against any attacks. The host signal or input signal is transformed into the feature signal. In this transformation, the normalized correlation is calculated between the host signal and the random signal. Then the modulation is performed on the feature signal and watermark message. The LSB method is applied in this modulated signal. Then CDMA method is applied. Then the embedding function is performed to produce the watermarked signal. This watermarked signal goes through the channel. At decoder side, the received signal is transformed into feature signal by using NCQIM technique which is done at embedded side and LSB method is used. Then the watermarked message is extracted from feature signal.

3.1 Normalized Correlation Based Quantization Index Modulation:

In the design methodology of the feature modulation watermarking method, a serious issue is to generate the feature signal from the host signal. For this purpose, the features should be selected by taking into account the properties of the utilized modulation technique. The feature signals are obtained by using PRNG seeded with the secret key. The host feature signal ([f.sub.x]) by computing the normalized correlation of the host vector (x) and a random vector (u) is,

[f.sub.x] = [x.sup.T]u/[parallel]x[parallel] [parallel]u[parallel] (1)

PRNG is an algorithm for creating a sequence of numbers whose properties almost equal to the properties of sequences of random numbers. It is completely calculated by a small set of initial values called the PRNG's seed. Pseudo random numbers are created by providing a random seed value to PRNG. Normalization is a process that changes the range of pixel intensity values. In watermarking, the normalized correlation is calculated between the host signal and random signal. The random signals are made by a PRNG seeded with the secret key, and also used for the generation of host signals and watermarked signals. This increases the probability of correct detection.

3.2 quantization index modulation:

The QIM used to embedding information by modulating a set of indices with the embedded information and then quantizing the input value with the related sequence of quantizes. In this method the amplitude of single pixel (or) a vector of pixels is quantized using quantizes. Given the quantization step A and a dither value d [epsilon] [[DELTA]/4, [DELTA]/4], the sets [[conjunction].sub.0] and [[conjunction].sub.1] are respectively represented by,

[[conjunction].sub.0] = {k[DELTA] + d/k [epsilon] z} [intersection] [-1, 1] (2)

[[conjunction].sub.0] = (k[DELTA] + d + 1/2[DELTA]\k [epsilon] z} [intersection] [-1, 1] (3)

This is used for normalized correlation. The normalized correlation [f.sub.x] is then modulated with the hidden symbol m by selecting a code word fm from the corresponding codebook [[conjunction].sub.m] to replace [f.sub.x.]. This can be implemented by a quantization operation.

3.2.1 Lsb Based Watermarking:

In this technique, the watermarking process is done in two methods: Watermarking Embedding techniques and Watermark Recovery techniques. In watermarking embedding the images are process in pixels and generate the watermark to cover original signal. In the embedding process, the size of the original image is obtained. Then the maximum message size to be embedded is determined based on the original image. Next the size of the key image is calculated. Then the watermark image is reshaped into vector. After this, the size of the watermark image is smaller than that of cover image. The watermark image is embedded into the cover image using LSB. Watermark extraction is done by the reverse process.

3.2.1 Cdma Based Watermarking:

In this, we propose a CDMA based multiple-user sharing-channel watermarking technique which is useful for applications in multiple-user sharing copyright protection. Although there are many watermarking algorithms related to CDMA, these single watermarking algorithms only utilize the spread spectrum characteristic of CDMA to enhance the robustness of the algorithms and do not make full use of its potential peculiarity of supporting multiple-user sharing channel.

[FIGURE 1 OMITTED]

To detect the watermark, each seed is used to generate its PN sequence, which is then correlated with the complete image. If the correlation level is high, that bit in the watermark is set to "1", otherwise a "0". The same method is then repeated for all the values of the watermark. CDMA improves on the robustness of the watermark significantly, but requires several orders more of calculation. In this, a novel approach based on the unconventional Random Encoding SS method is proposed for recovering the spreading sequence of watermark signal without any information from the transmitter. It is contribute a higher secure feature by using of the time varying random encoded spread spectrum.

3.3 Watermark Embedding And Extraction Function:

The embedding function Em (.) is designed to produce a watermarked signal. In this model, the information modulation is not directly done on the input signal itself, but based on a feature signal transformed method from the host signal, which is quite different from the existing watermarking methods. The performance improvement can be brought about if an appropriate feature is utilized. The watermark signal is embedded in the original image after applying LSB technique. The watermarked signal goes through a particular channel, where it might be exposed to several common signal processing manipulations. The Gaussian noise is added in the channel to reduce the losses. This Gaussian noise is added in the channel to produce distorted watermarked signal.

The distorted watermarked signal can be written as,

z = y + n (4)

Where, z - Distorted watermarked signal, Y - Watermarked signal,

3.4 Watermark Extraction:

At the decoder side, the received signal z is first transformed into the feature signal [f.sub.z] as done at the embedded side. Then, a message [??] is extracted from [f.sub.z] according to the correspondence between the watermark content and the used the same feature level, which is established during the information modulation process and a message is extracted from [f.sub.z].

4 Parameter Analysis:

4.2 Watermark Imperceptibility:

Imperceptibility means that the watermarked content is perceptually indistinguishable from the original image. This property is related to the robustness of the watermark and hence an optimal balance between imperceptibility and robustness must be achieved by the watermarking scheme. Imperceptibility can be measured using Peak Signal to Noise Ratio (PSNR) between host and watermarked image and Root Mean Squared Error (RMSE). The PSNR value is calculated by using Mean Squared Error (MSE).

4.2.1 Peak Signal to Noise Ratio:

The visual quality of the embedded images can also be measured using the PSNR. It is the most easily defined by the Mean Squared Error (MSE). The PSNR is defined as

PSNR = 10[log.sub.10] ([MAX.sup.2]/MSE) in decibels (5)

Where, MAX--Maximum pixel value of the frame, MSE--Mean Squared Error

The MSE measures the statistical difference in the pixel values between the original and the reconstructed image. The MSE is defined as,

MSE(x,y) = 1/MN [[summation].sup.M.sub.y=1] [[summation].sup.N.sub.x=1] ][I(x,y) [I.sup.I] (x, y)].sup.2] (6)

Where,

X--Original image, Y--Watermarked image, M, N--Size of the original image.

The RMSE is a quadratic scoring rule which measures the average magnitude of the error. It is calculated by using the formula,

RMSE = sqrt(MSE) (7)

4.3 Robustness:

The robustness can be achieved by using the parameter BER. Bit Error Rate is determined by Inverse of PSNR values.

BER = 1/PSNR (8)

The more is BER lesser will be the quality of the Watermarking technique. So BER is inversely proportional to PSNR. More the value of BER lesser will be PSNR value.

5 Result Analysis:

In this paper the simulation process is implemented in MATLAB using different types of host images and watermark images. For analysis purpose gray scaled image is used. The opted host (original) and watermark images are as follows: The performance evaluation can be analyzed by using the Peak Signal to Noise Ratio (PSNR). The following figures represent the extracted watermark images with several attacks of the proposed scheme. The different levels of values of PSNR & MSE for different images are listed in Table 2 and 3.

[TABLE 1 OMITTED]

Also for different images the values of RMSE and Bit Error Rate also changes. The different values of Root Mean Square Error and Bit Error Rate for different images have been listed in Table 4 and 5.

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

[FIGURE 5 OMITTED]

Conclusion:

In this paper, a novel quantization based watermarking method, called NCQIM based LSB combined with CDMA technique, has been developed. In the method, the NC between the host signal and a random signal was modulated with the hidden message using the structured codebooks, and then the watermarked signal was formed to provide the modulated NC by minimizing the embedding distortion. The induced watermark signal by NCQIM is a linear combination of the host signal and the random signal, which is perceptually better than random watermark signals independent of the host signal. Moreover, we theoretically evaluated the performance of the proposed method. According to our analyses, the embedding distortion of NCQIM is a monotonically decreasing function of the quantization step and proportional to the host signal variance. NCQIM has been also extensively tested on various images and compared with the state-of-the-art watermarking methods. Experimental results demonstrate that our method obtains the improved imperceptibility and a relatively high embedding capacity in a very noisy environment. Moreover, it yields superior robustness against a wide range of common image processing operations. This proposed method further can be extended with various embedding algorithm to enhance the robustness level.

REFERENCES

[1.] Xinshan Zhu, Jie Ding, and Honghui Dong, 2014. "Normalized Correlation-Based Quantization Modulation For Robust Watermarking" IEEE Transactions On Multimedia, 16: 7.

[2.] Zhuand, X., J. Ding, 2012. "A novel quantization watermarking scheme using random normalized correlation modulation," Chin.J.Comput., 35(9): 1959-1970.

[3.] Zhu, X. and S. Peng, 2012. "A novel quantization watermarking scheme by modulating the normalized correlation," in Proc. Int. Conf. Acoust., Speech Signal Process., pp: 1765-1768.

[4.] Guerrini, F., M. Okuda, N. Adami and R. Leonardi, 2011. "High dynamic range image watermarking robust against tone-mapping operators," IEEE Trans. Inf. Forensics Security, 6(2): 283-295.

[5.] Kalantari, N.K. and S.M. Ahadi, 2010. "A logarithmic quantization index modulation for perceptually better data hiding," IEEE Trans. Image Process., 19(6): 1504-1517.

[6.] Kalantari, N.K., S.M. Ahadi and M. Vafadust, 2010. "A robust image watermarking in the ridge let domain using universally optimum decoder," IEEE Trans. Circuits Syst. Video Technol., 20(3): 396-406.

[7.] Mankar, V.H., T.S. Das, and S.K. Sarkar, 2009. "An angle QIM watermarking in STDM framework robust against amplitude scaling distortions," Contemporary Comput., Commun. Comput. Inf. Sci., 40(8): 400-410.

[8.] Cox, I.J., M. Miller, J.A. Bloom, J. Fridrich and T. Kalker, 2008. Digital Watermarking and Steganography, 2nd ed. San Francisco, CA, USA: Morgan Kaufmann.

[9.] Comesana, P. and F. Perez-Gonzalez, 2007. "On a watermarking scheme in the logarithmic domain and its perceptual advantages," in Proc. IEEE Int. Conf. Image Process., San Antonio, TX, USA, pp: 145-148.

[10.] Li, Q. and I.J. Cox, 2007. "Using perceptual models to improve fidelity and provide resistance to valumetric scaling for quantization index modulation watermarking, " IEEE Trans. Inf. Forensics Security, 2(2): 127-139.

[11.] Lan, T.H., A.H. Tewfik, 2006. "A Novel High-Capacity Data- Embedding System," IEEE Trans. On Image Processing, 15(8): 2431-2440.

[12.] Perez-Gonzalez, F., C. Mosquera, M. Barni and A. Abrardo, 2005. "Rational dither modulation: A high-rate data-hiding method invariant to gain attacks," IEEE Trans. Signal Process., 53(10): 3960-3975.

[13.] Ourique, F., V. Licks, R.. Jordan and F. Perez-Gonzalez, 2005. "Angle qim: A novel watermark embedding scheme robust against amplitude scaling distortions," in Proc. Int. Conf. Acoust., Speech Signal Process, pp: 797-800.

[14.] Moulin, P. and R. Kotter, 2005. "Data-Hiding Codes," Proceedings IEEE, 93(12): 2083-2127.

[15.] Celik, M., G. Sharma, A. Tekalp and E. Saber, 2005. "Lossless generalized-LSB data embedding," IEEE Trans. Image Process., 14(2): 253-26.

[16.] Balado, F., 2005. "New geometric analysis of spread-spectrum data hiding with repetition coding, with implications for side-informed schemes," in Proc.Int.Workshop Digital Watermark., pp: 336-350.

[17.] Bartolini, F., M. Barni and A. Piva, 2004. "Performance analysis of ST-DM watermarking in presence of non-additive attacks," IEEE Trans. Signal Process, 52(10): 2965-2974.

[18.] Wu, M. and B. Liu, 2003. "Data hiding in image and video: Part I-fundamental issues and solutions," IEEE Trans. Image Process., 12(6): 685-695.

[19.] Perez-Gonzalez, F., F. Balado and J.R.H. Martin, 2003. "Performance analysis of existing and new methods for data hiding with known-host information in additive channels," IEEE Trans.Signal Process., 51(4): 960-980.

[20.] Eggers, J.J., R. Bauml, R. Tzschoppe and B. Girod, 2003. "Scalar Costa scheme for information embedding," IEEE Trans. Signal Process, 51(4): 1003-1019.

[21.] Fridrich, J., M. Goljan and R. Du, 2002. "Lossless data embedding-new paradigm in digital watermarking," EURASIP J. Appl. Signal Process., 1: 185-196.

[22.] Chen, B. and G.W. Wornell, 2001. "Quantization index modulation: A class of provably good methods for digital watermarking and information embedding," IEEE Trans. Inf. Theory, 47(4): 1423-1443.

[23.] Chen, B. and G.W. Wornell, 2001. "Implementations of quantization index modulation methods for digital watermarking and information embedding of multimedia," J. VLSI Signal Processing Syst. Signal, Image, and Video, Technology (Special Issue on Multimedia Signal Processing), 27: 7-33.

[24.] Digital Image Processing using Matlab By, Gonzalez

[25.] Cox, I.J., M.L. Miller and A. McKellips, 1999. "Watermarking as communications with side information," Proc. IEEE, 87(7): 1127-1141.

[26.] Deller, J.R., Jr., J.G. Proakis and J.H.L. Hansen, 1993. Discrete Time Processing of Speech Signals, New York: Macmillan.

(1) M. Subbulakshmi and (2) M. Vijayaraj

(1) Department of ECE, National College of Engineering, Maruthakulam-627151, Tamilnadu, India.

(2) Department of ECE, Government College of Engineering, Tirunelveli-627007, Tamilnadu, India.

Received 15 May 2016; Accepted 7 July 2016; Available 22 July 2016

Address For Correspondence:

M. Subbulakshmi, Department of ECE, National College of Engineering, Maruthakulam-627151, Tamilnadu, India.

E-mail: subbu_june23@yahoo.co.in

Table 2: Comparison of PSNR value Original Watermark PSNR value S. no Images Images of NCQIM 1 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 44.9600 2 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 48.0021 3 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 45.4116 4 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 45.4745 5 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 53.5258 6 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 45.9297 7 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 48.7510 8 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 45.9314 9 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 50.9812 10 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 49.2315 PSNR value using PSNR value of NCQIM, LSB, S. no NCQIM, LSB CDMA 1 45.6650 48.3650 2 48.6051 55.8901 3 45.9504 49.6891 4 45.9944 49.3075 5 53.7265 55.2681 6 46.2213 47.7285 7 49.2098 50.1713 8 46.1259 48.6234 9 51.7258 52.3651 10 49.7821 50.7268 Table 3: Comparison of Mean Squared Error S. no Original Images Watermark Images 1 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 2 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 3 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 4 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 5 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 6 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 7 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 8 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 9 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 10 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] MSE value of MSE value of NCQIM, S. no MSE value of NCQIM NCQIM ,LSB LSB, CDMA 1 2.09 1.78 1.10 2 1.03 0.89 0.18 3 1.88 1.66 0.89 4 1.86 1.65 0.90 5 0.28 0.27 0.19 6 1.66 1.55 1.20 7 0.87 0.78 0.71 8 1.65 1.59 0.98 9 0.52 0.43 0.37 10 0.77 0.67 0.53 Table 4: Comparison of Root Mean Squared Error S. no Original Images Watermark Images 1 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 2 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 3 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 4 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 5 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 6 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 7 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 8 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 9 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 10 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] RMSE value of RMSE value of RMSE value of S. no NCQIM NCQIM, LSB NCQIM, LSB, CDMA 1 1.4457 1.3342 1.0488 2 1.0149 0.9434 0.4242 3 1.3711 1.2884 0.9433 4 1.3638 1.2845 0.9486 5 0.5292 0.5196 0.4358 6 1.2884 1.2449 1.0954 7 0.9327 0.8832 0.8426 8 1.2845 1.2609 0.9899 9 0.7211 0.6557 0.6082 10 0.8774 0.8185 0.7280 Table 5: Comparison of Bit Error Rate S. no Original Images Watermark Images 1 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 2 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 3 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 4 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 5 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 6 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 7 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 8 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 9 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] 10 [ILLUSTRATION OMITTED] [ILLUSTRATION OMITTED] BER value of BER value of BER value of S. no NCQIM NCQIM, LSB NCQIM, LSB, CDMA 1 0.0222 0.0219 0.0206 2 0.0208 0.0206 0.0178 3 0.0220 0.0218 0.0201 4 0.0219 0.0217 0.0202 5 0.0187 0.0186 0.0180 6 0.0218 0.0216 0.0214 7 0.0205 0.0203 0.0199 8 0.0218 0.0217 0.0205 9 0.0196 0.0193 0.0190 10 0.0203 0.0200 0.0197

Printer friendly Cite/link Email Feedback | |

Author: | Subbulakshmi, M.; Vijayaraj, M. |
---|---|

Publication: | Advances in Natural and Applied Sciences |

Date: | Jul 1, 2016 |

Words: | 4414 |

Previous Article: | Optimizing multi GPU with fixation based sculpture restoration on pixel-by-pixel using deep convolution neural network. |

Next Article: | Thermal buckling for nano-composite plate. |

Topics: |