A Novel Hybrid Watermarking scheme with Image authentication based on frequency domain, 2- Level SVD using chaotic map

INTRODUCTION: In order to protect the owner’s visual multimedia data, various watermarking methods have introduced in the last few years. In image watermarking techniques, a secret watermark image inserted into another image. OBJECTIVES: The main aim of this paper is to implement an image watermarking technique having high security from stealing the information hidden in image and make the resultant image more robust against various attacks. METHODS: In this paper, frequency domain based techniques like DWT, DCT has used with 2 Level SVD and a new 3D discrete hyper chaotic map. The digital signature and other image parameters have appends in the watermark image to ensure the security and copyright protection of the original image using RSA, Arnold transforms, and SHA-1, techniques. At the receiver side, digital signature and information parameters have retrieved from the watermark image and identify to detect whether an unauthorized person attacks the watermark image or not. To verify the originality of image, the extracted information parameter, and the hash value of the extracted watermark have compared with the original value of the parameters and source hash value of the watermark image, respectively. The secret key used in the proposed scheme that makes the system more robust and adds security to the system. It helps to authenticate the originality of the received image. RESULTS: We compare the result with the present scheme proposed by Lin at el. The experiments have conducted to check various standard image processing attacks, and the watermark extraction process generates a high-quality image after different attacks. The primary purpose of the suggested watermarking scheme is to ensure reversibility and high security. CONCLUSION: The experimental outcomes confirmed that not only the recommended scheme attained higher robustness and imperceptibility in comparison to our earlier scheme. The Watermark image also recovered successfully with the known secret key.


Introduction
Today the Internet has transformed the whole world into a global village of sharing and transferring digital data such as video, audio, images, etc. [1]. At the same time it creates illegal duplicate copies by hackers or unauthorized persons that seriously affect the author's copyright. The resulting problems cause copyright protection and security [2]. Therefore, how to avoid copyright violations and protect intellectual property has attracted worldwide attention from some researchers. The varieties of methods used for the purpose are cryptography, steganography and watermark.
To handle the first issue of copyright protection, Information Hiding Techniques (IHT) is necessary [23]. Generally, the IHT should satisfy four basic requirements such as volume, security, robustness, and noiselessness. But normal IHTs do not meet all these requirements, cause a tradeoff relationship. The embedding of a large amount of message inside the digital image cause apparent artifact, but enhancing the noiselessness will reduce the volume of the embedding message. Hence, needs of suitable information (data) hiding techniques based on the applications [4] is required. According to that, the IHTs are categorized into two groups of Transform Domain (TD) and Spatial Domain (SD) [17]. In SD, a direct modification is performed inside the content value of the pixel of the input image during message embedding. The following types of techniques used are Least Significant Bit Insertion (LSBI), patchwork, and vector quantization. The advantage of SD has a higher embedding capacity, and it has lower computational complexity. But in TD, the message embedding is performed by modulating the transform domain coefficients such as Discrete Cosine Transformer (DCT) [32], Lifting Wavelet Transform (LWT), Discrete Fourier Transformer (DFT), Discrete Wavelet Transformer (DWT) and Singular value Decomposition (SVD) [7,9]. They can be applied either separately or together according to the need of requirements. Recently, reversible data embedding techniques are introduced [8].
Considering the second issue of security in the information system, proper image identity assurance and integrity is necessary [22]. The image integrity means identifying the received image is a correct image from an accurate source. Image integrity involves making a confirmation on original images that no changes have been made during processing and transferring acquired from a medical imaging device. To undergo identity and integrity security issues, Digital Image Watermarking (DIW) techniques are introduced for images for copy right protection [26]. DIW embeds the secret information in the form of a digital image, hence it is difficult to remove or change the content from a digital image. According to their robustness, they are categorized into fragile, semi-fragile, and robust techniques. On the receiver side the hidden image is retrieved by the watermark extraction process. The tamper detection of digital data and copy control is considered as the main objective in digital watermarking to be achieved. The requirement should be consider while performing good watermarking technique is, the embedded information do not degrade the quality of the host image [28].
Sometimes, information present in an image is in nonuniform nature and distributed across the images. Some of the parts include more information than other parts. Thus several schemes are utilized in separating various regions and objects in medical images. According to that, the medical image has split into two areas, such as the Area of Interest (AOI) and Area of Non-Interest (AONI) [19,16]. The two domains utilized for inserting the watermark that contains secure information in the image is Pixel Domain (PD) and Coefficient Domain (CD). In PD original image pixels are directly altered with the watermark bits. This method is easy to implement, produces less computational cost and has a high payload, but most of the PD offers the least robustness. In CD, the frequency coefficient of any transform is altered based on watermark bits. Where DCT and DWT are the most common frequency domain transform techniques used in the CD watermark method [21]. The PD and CD also called SD and TD. Performing digital watermark in SD, the data is directly embedded in pixels in replace of the pixel bits into data bits. But it is not robust when a high payload is required and produces lossless compression. But performing digital watermark in TD, the coefficients of the cover medium are highly modified offers more security and high robustness during inner and outside attacks [15].
The main aim is, to make the algorithm robust and secure. In this paper, a secure hybrid and robust image watermarking (RIW) scheme based on DWT, DCT, 2-Level SVD and a new 3D discrete hyper chaotic map for the information hiding technique is proposed.
The paper is organized in different sections. In section 2 literature survey has done. In section 3 preliminaries, a brief overview of the above terms is discussed. In section 4, the proposed schemes, tools, and techniques used have discussed. In the next section 5, the experiment result has discussed. In the last section, conclusions have discussed.

Literature Review
Some of several encryptions transform domain and steganography techniques used in watermarking to improve the security and robustness is given below.
Hurrah et al. [1] aimed a hybrid transformation domain watermark scheme for an efficient watermark. The main goal is to resist dual attacks of image processing attacks and geometric attacks. For this, the author uses multiple encryption techniques to increase the security of the information content present in the watermark image and used a principle of maximum probability to select the bits of encrypted watermark for a color image. The resultant evidence is highly robust in image processing attacks not for geometric attacks.
Anand et al. [29] proposed a robust and secure watermarking technique. In this paper secure patient data is embedded using DWT-SVD techniques in medical image then the resultant image is encrypted and compressed. Finally encrypted image is transmitted for communication channel.
Ye, et al. [30] propose a new image encryption algorithm based on compressive sensing and information hiding technology using DWT. In this logistic-tent map is used for creating randomness in image pixel. Ramasamy, et al. [31] proposed enhanced logistic map to encrypt the image using modified zigzag transformation. The results show that system is secure, reliable and good efficiency.
Chen and Xu [6] proposed a robust and secure color image watermark scheme for copyright protection based on the LWT domain to decompose the watermark and A Novel Hybrid Watermarking scheme with Image authentication based on frequency domain, 2-Level SVD using chaotic map host image into multiple sub-bands. The text and watermark images used are RGB color images. To demonstrate the solidity of the scheme in the watermark image various medium filtering image processing attacks, cropping attacks, etc. are applied. But in this method, only the original image is calculated for imperceptibility, do not consider volume and robustness.
Abdel hakim et al. [5] proposed a Time Efficient Optimization (TEO) method based on Machine Learning (ML) approach. The aim of this method is to find bestembedded strength parameters for RIG technique. Initially, they designed the DCT domain to find the common watermarking attacks. Then applied a watermarking to train the image and used an Artificial Bee Colony (ABC) algorithm to optimize the parameters to extract the features of the image with best fitness value. Finally, the optimum embedded parameters are predicted by the K-Nearest Neighborhood regression method.
Loan et al. [10] presented a DMI technique with Chaotic Encryption (CE) for both color and gray scale images [20]. Before embedding the secured watermark in the input image, the DCT coefficient operation has performed to avoid the overlapping problem. Then double-layer security is used in the embedding stage using the Arnold transformation in CE. The result shows better achievement in terms of security, noiselessness, and robustness.
Singh AK [11] used three TD techniques of DCT, DWT, and SVD for hybrid watermarking. In the initial stage, the S vector of the watermark information inserted in the S segment of the source image. The generated watermark is performed by inverse Singular value decomposition in a modification of S vector and by U, V vector by inverse DWT and DCT. An extraction algorithm extracts the secured watermark. Finally, an encryption method is embedded into the cover image to perform security and reduce the storage capacity and bandwidth in the source image [33].
Yahya et al. [12] explain a robust watermarking algorithm which is based on the Probabilistic Neural Network (PNN) in the DWT domain with the Haar filter. The Haar filter is used to incorporate the image of the binary watermark into the corresponding coefficient wavelet blocks. To extract the image of the watermark, PNN is applied between the watermarks to evaluate the efficiency and quality of the image. Finally, the demonstration is performed in Normalized Cross Correlation (NCC), Peak to Noise Ratio (PSNR) and some common attacks.
Chen et al. [13] designed a scheme based on DWT in which logo is verified without using original host image. Sanjay Rawat and Balasubramaniam Raman [3,14] proposed a scheme using discrete cosine transform and logistic regression on grey scale image without the presence of host image for logo verification [24,25]. The secret key is obtained using the original source image but can't be easily seen by human eyes. This theory was applicable only for grey scale image but not for the color image. This paper presents the application on color host image and color watermark embedding / extraction method.
In the different watermarking schemes, some use secret text images to be hidden in digital images and finally encrypted the image to obtain an encrypted watermarked image. Many algorithms use a different encryption algorithm to encrypt the watermarked image. In our paper, we have the first encrypted watermark logo with some additional information on it. And embed encrypted watermark image into source image using a new proposed chaotic map. The new chaotic map is hyperchaotic and very sensitive to the initial value of the map as discussed in section 3, which adds security in the system that prevents sensitive information from the unauthorized person.In this paper, hybrid digital watermarking is carried out using DWT, DCT, 2-Level SVD and new chaotic map which is a novel scheme for image has discussed.

Preliminaries
In this preliminaries section, we explain the DCT, DWT, SVD, SHA-1, RSA, Arnold transform and proposed 3D discrete hyper chaotic map requisite needed in our proposed technique.

2D Discrete Cosine Transform (DCT)
The transformation is a major component in the processing of multimedia data applications. In the transformation for an image, pixels display some correlation with their neighbouring pixels in a single image. Also, pixel value is easily calculated from its nearest neighbours by using this correlation. Therefore, a transformation of pixels shows how correlated coefficients maps into uncorrelated coefficients [18].
The DCT converts/transforms an image pixels or the signal values from the spatial (correlated) to the frequency domain (uncorrelated).
The inverse cosine tranform for the aboce eqaution is: Here f (i, j) is pixel without transformation and F (u, v) is pixel after transformation. M and N are block size for DCT.
where F(0,0) in eq. 3 is DC coefficients and others are AC coeeficients.

Discrete Wavelet Transform
The wavelet transform of signal done by discrete wavelet transform (DWT) using distinct sets. It has widely used in image watermarking techniques. The DWT decomposes an image into the frequency domain having frequency sub-bands. DWT is the most common frequency domain transform technique used in the coefficient domain. It converts the signal into two parts; one part contains the high-frequency component (HFC), and the other includes the low-frequency component (LFC). The information about the edge component of the image is contained in the high-frequency component. And low-frequency component is further decomposed into low and highfrequency components. The human eyes are sensitive to changes made in non-edge portions. Usually, HFC is suitable for watermarking techniques because changes done in HFC are indistinguishable by the human eye. In an image processing first DWT is applied in a vertical direction than in the horizontal direction. After using the DWT image is converted into the frequency domain, it contains four frequency sub-bands named as High High (HH), Low High (LH), Low Low (LL), and High Low (HL). The LL frequency sub-bands has used as input for further decomposition.

Singular Value Decomposition (SVD)
Singular value decomposition (SVD) is a standard transform applied in various numerical analysis. By using SVD, a matrix could be decomposed into eigenvalues and eigenvectors. For a given matrix, the SVD can be represented as given below: In equation 4, 'V' represents the right singular vectors and 'U' represents left singular vectors. S matrix are called diagonal matrix. S is represented as: S = diagonal ( λ 1, λ 2 , λ 3, …. , λ r) where λ 1 ≥ λ 2 ≥ λ 3………….≥ λ r are called the singular values of A matrix. The singular values have high energetic parts and produce excellent stability. It can resist the slight irregularities in image processing. Also, the SVD can be applied to any arbitrary matrix [34]. In addition to that, the SVD transform technique has widely utilized in various robust watermarking schemes.

SHA-1 Hash Generation
SHA-1 is an acronym as a "Secure Hashing Algorithm." It is proposed by the US National Security Agency (USNSA) and issued by NIST. It is a hash algorithm and has better performance than the previous SHA-0 algorithm. It becomes more popular after published in 1995. It produces the 160 bits of message digest size. Nowadays, there are various versions of secure hash function has implemented like SHA-224, SHA-256, SHA-384, and SHA-512. SHA-512 hashing uses 512bits for message digest size, and block size is 1024bits. In this paper, SHA-1 hashing technique has used, which generates 160bits for any input or file. For example, suppose the message is "hello word" then its hash value using SHA-1 hashing algorithm the calculate value has represented as "e0738b87e67bbfc9c5b77556665064446430e81c".

RSA
RSA algorithm is a public-key cryptography technique. It is broadly useful in various secure information transmission channels. RSA's name derived from the three scientists named Ron Rivest, Adi Shamir, and Leonard Adleman. They introduced the RSA encryption algorithm in 1978. It is an asymmetric cryptographic algorithm that uses two keys i.e., private and public key. In the RSA encryption algorithm, the public and private keys collectively encrypt a message in plain text, and the different key is a key used to decrypt a message which is encrypted. It ensures data integrity, authenticity, and confidentiality. The RSA bits may be typically 1024 or 2048bits, and the key derived from the large prime numbers.

Arnold Transform
Arnold transform is a simple and most common encryption technique widely adopted to hide sensitive information from the external world i.e., from unauthorized persons. The general form of Arnold transform can be represented as given below: Where (xi, yi) is the original coordinate of image pixel; (xi+1, yi+1) is the scrambled coordinate, and W denotes the width of the input image. The Arnold conversion has the periodicity property, which means that after a certain amount of iterations, the original image will reappear. The numbers of times of iteration act as a secret key for Arnold transform. For secure transmission of watermark, the Arnold transform is performed on the watermark to get encrypted watermark that can be inserted securely into the source image. Without the knowledge of secret key watermark cannot be decrypted.

3D Discrete Hyper Chaotic Map
In this paper, a new 3D discrete hyper chaotic map with strong chaotic behavior is proposed for digital watermarking scheme. The new 3-dimensional discrete cosine hyper chaotic (HCM) map is governed by the following equation. Where x0, y0, and z0 are initial conditions. The variables a and b are the control parameter of the HCM map in equation (6). The variable n is the number of times to iterate HCM map. In experiments the initial value of x0, y0, z0, a and b are taken as 1.123, 0.479, 2.317, 7.123 and 5.7 respectively. The sensitivity to initial condition of any chaotic map or system is quantified by Lyapunov Exponent (LE) metric.
LE is the rate of separation between two close trajectories The System is said to have chaotic behavior if one LE is positive and if it has at-least two positive LEs, then system is hyper-chaotic The LE diagram of HCM map versus parameter a and time are shown in Figs. 2(a) and 3(b) respectively. The Bifurcation plot and spatiotemporal diagram of HCM maps is presented in Figs. 2(b) and 3(a), respectively.

Fig 2. (a) Lyapunov exponent diagram versus parameter a, (b) Spatiotemporal diagram with initial values
A Novel Hybrid Watermarking scheme with Image authentication based on frequency domain, 2-Level SVD using chaotic map

Proposed Method
In this part, we describe the steps of proposed method. Firstly, the watermark image encrypted using the encryption process before the embedding the watermark.
The block diagram of the encryption algorithm shown in fig 6(a).In the encryption process, first, we generate a hash value of the input watermark. Then apply the RSA encryption algorithm on the generated hash content to obtain the digital signature (DS).We append some information parameter to the watermark image with DS. Then Arnold transform has applied to obtain encrypted watermark image (EWI) as shown in fig 6(a). The information parameter includes m, n is the dimension of source image, p, q is the dimension of watermark image and 'a' is the rotational angle of the source image. The angle is calculated by using the corner pixel coordinate of the image [27].
At the received end, inverse Arnold transform has applied to the EWI image. After that, we extract watermark, DS, and information parameters from the resultant image, as shown in fig 6(b).We checked the extracted parameter value from source image characteristics. Let extracted parameter are m', n,' p,' q,' and a.' If the dimension has differed, then the image is altered by the unauthorized person. The angle has calculated by the value of corner pixel of received image. If the amount of the angle is zero, then there is rotation attack on received image otherwise image has rotation attack.

Fig 4. The block diagram of embedding process
In the encryption process of the JMI logo, number of pixel of change rate (NPCR) value is 99.455, the unified averaged changed intensity (UACI) is 33.451 and entropy is 7.939. The adjacent pixels correlation in encrypted images is 0.00549. For copyright logo encrypted image NPCR, UACI, entropy and adjacent pixels correlation are 99.639, 36.48, 7.986 and 0.00637 respectively.
Here, we explain the proposed algorithm for image watermarking technique. We take image I as input image and the size of image is 'm x n' and W notation is used for watermark image. The dimension of the watermark is 'p x q'. The proposed embedding and extraction watermark scheme methodology have shown in the block diagram in Fig 4 and Fig 5,

The Proposed steps of the shuffling algorithm
The shuffle algorithm takes matrix as input and shuffle the matrix using HCM map. The initial value used in HCM map is mentioned in 3.7 section. The fig 7. show the steps of the shuffling algorithm.
Where Wi, j is pixel value of embedded watermark and Wi, j″ is the pixel value of extracted watermark respectively; m × n is image size. The cumulative NC value is calculated by taking the maximum of NC values of Red, Blue and Green components. The effectiveness of the technique is tested under different common image processing attacks like rotation, blurring, noise addition, compression etc. The 'Lena' image is gone under different common image processing attacks and the visual results are shown in table 1.
The Structure Similarity Index Method (SSIM) is used to quantify image quality caused by various common image processing attacks. It is measure the similarity between the images. For calculating the performance evaluation of the proposed technique, we compare the result with the present scheme proposed by Lin and Lin at al. [4]. Table 2 shows the comparison result with the existing scheme. The NC value is high in most of the cases. Table 3 shows the PSNR result of JMI logo watermark. In table 4 we have shown the result of copyright logo watermark result. The outcomes show that the proposed technique of watermarking is robust and has a strong capability to the different attacks. It also checks the authenticity of the image at the receiver side.

Conclusion
In this paper, the image authenticated at receiver end by using information parameter and digital signature. To add more security in the watermarking scheme, the feature of the DCT, DWT, HCM map, and 2 Level SVD has effectively used. The experimental outcomes confirmed that not only the recommended scheme attained higher robustness and imperceptibility in comparison to our earlier scheme. The Watermark image also recovered successfully with the known secret key. Without the information of the private key, the embedded watermark cannot be retrieved.