1 Introduction

Transmission of multimedia data such as images, videos and audios has significantly increased in the last decade. However, such a massive usage of the Internet also provided some opportunities for the eavesdropper. Eavesdropper apply different attacks and can access private and secret information. Therefore, to secure the information from illegal access and distribution, information hiding has recently become a major area of research. Intruders apply various attacks such as changing meaningful information through modifying the content or decipher the actual information. Steganography, cryptography, and watermarking have one thing in common: they all transmit secret information in such a way that only the receiver can decrypt. Steganography and cryptography are two separate approaches for keeping data secure. The goal of steganography is to conceal hidden messages in media files in such a way that no one can detect the existence of a secret message. A watermark is a type of marking that is hidden within a noise-tolerant signal like audio, video, or image data. It is commonly used to determine who owns the copyright to a signal with the main purpose of watermarking is to ensure proof of ownership. In watermarking, secret data is embedded in the original image in the form of text or image to shield the ownership of original image against unlawful copies. Watermarking can be mainly classified into blind, semi-blind and non-blind watermarking. Blind watermarking extracts the watermark without any knowledge of the original and watermarked images. The semi-blind watermarking extracts the original image knowing the watermarked image while the non-blind watermarking extracts the original image knowing original image, watermarked image and watermark [5, 20].

Over the last decade, various watermarking schemes have been proposed [6, 7, 9, 11, 12, 21, 23, 25, 28, 31, 32, 34, 35, 37]. For instance, a secure and Singular Value Decomposition (SVD) based watermarking method was proposed in [7, 25]. However, due to greater complexity and higher computational complexity, SVD based watermarking schemes are not suitable for hardware implementation. In [11, 23, 28], authors modified Discrete Cosine Transform (DCT) coefficients to embed the watermark. In 2012, Vahedi et al. improved the performance of some existing schemes by utilizing genetic algorithm optimization principles and proposed a new watermarking scheme for colour images [37]. Das et al. utilized the correlation among DCT adjacent coefficients and designed a blind efficient image watermarking scheme in the transform domain [12]. Authors in [8, 22] found that using 8 \(\times\) 8 block-wise DCT based watermarking schemes can decrease real-time hardware implementation problem. In [26, 27], Moulin et al. and Mukherjee et al. designed secure watermarking schemes, respectively. Authors in [24], presented sub-sampling and Discrete Wavelet Transform (DWT) based watermarking scheme in the frequency domain. The proposed scheme in [24] resist geometric distortions. Zhou et al. inserted watermark in mid frequencies of Fourier Mellin Transform (FMT) for the minimization of the geometric deformation caused by lawful printing and scanning processes [41].

Since the last decade, a close relationship between chaos and cryptography has been proved. Due to the initial key sensitivity, non-periodicity, deterministic pseudo randomness, and larger keyspace, chaos-based watermarking schemes have received a lot of attention from the research community. In literature, a number of chaos-based watermarking schemes are available [13,14,15,16, 18, 19, 29, 36, 40]. The most commonly used chaotic maps are Henon map, Logistic map, Arnold map, Lorenz equations, Skew Tent map, Standard map and Baker map. Dawei et al. designed a chaos-based watermarking scheme for still images by transforming the image into the frequency domain. Dawei et al. applied wavelet transformation locally and inserted the chaos-based watermark into sub-band coefficients [13]. Authors in [40] presented a new watermarking scheme using two chaotic maps. Embedded indexes of the host images were encrypted by one chaotic map while the second chaotic map was used to locate the pixel bit of the host image. Fan et al. proposed a novel watermarking scheme for speech data copyright protection via Discrete Fractional Sine Transform (DFRST) [14]. The security of the watermarking scheme [14] was improved through adopting chaotic sequences. In [29], a novel chaos-based fragile watermarking scheme for image authentication and tamper detection was presented. This watermarking scheme can identify the modification in an image and can also detect the exact location that has been modified. For security enhancement, two chaotic maps were employed in the presented algorithm [29].

Self-recovery and tamper detection fragile watermarking scheme were presented by Tong et al. [36]. Blocks of the original image were confused by using the cross chaotic map. The recovery after tampering was improved through a novel method known as sister block embedding scheme [36]. Ghebleh et al. proposed blind frequency domain watermarking scheme using Logistic and Arnold chaotic maps [15]. According to [15], the proposed wavelet transform based algorithm inserts a white and black watermark logo in the mid-band component of the host image. Two copies of watermark were embedded in two distinct subbands of the host image. The computational complexity was reduced by using only the least significant bits of the image [18]. Furthermore, Khan et al. [19] utilized classes of finite chain rings for designing of substitution Box (S-Box). The authors developed image encryption and watermarking schemes using the aforementioned S-Box. Jamal et al. embedded watermark in the frequency domain of the original image [16]. The embedded indexes were generated through random numbers that were obtained from Chaotic Fractional Rössler System [16]. Su et al. [33] utilized Hessenberg transform and designed a blind colour image watermarking algorithm. The security of the proposed watermarking scheme was increased via Arnold transform and MD5 hashing. Roy et al. developed a DCT-based blind colour watermarking scheme for inserting multiple watermarks simultaneously [30].

1.1 Contributions

Following are the main contributions of this paper.

  1. 1.

    For higher security, SHA-256 is applied on the original image and initial condition is generated for PWLCM chaotic map.

  2. 2.

    A dynamic \(16 \times 16\) S-Box is generated from the random data obtained using PWLCM. Due to the dynamic nature of substitution, the presented scheme resist a number of statistical attacks.

  3. 3.

    The final watermark image is obtained using dynamic S-Box and the HH and HL block of the lifting wavelet transformed image.

The rest of paper is structured as follows: Sect. 2 presents the fundamental knowledge. Section 3 discusses the proposed scheme. Sections 4 and 5 presents simulation results and conclusion, respectively.

Fig. 1
figure 1

Plots of PWLCM against number of iterations for slightly different values of \(z_{0}\)

Full size image
Fig. 2
figure 2

Distribution of z against iterations

Full size image

2 Fundamental Knowledge

2.1 PWLCM

Due to the random-like nature and sensitivity to initial conditions, a number of researchers have used PWLCM map in image encryption schemes [10]. From previous research [3, 38], it is evident that PWLCM has good dynamical behavior, uniform invariant density, and ergodicity. Mathematically, PWLCM map can be written as [39]:

$$\begin{aligned} \begin{array}{r} z=f\left( z_{m-1}, \alpha \right) \\ =\left\{ \begin{array}{cr} \frac{z_{m-1}}{\alpha }, &{} 0 \le z_{m-1}<\alpha \\ \frac{\left( z_{m-1}\right) }{(1-\alpha )}, &{} 0 \le z_{m-1}<0.5 \\ 0, &{} z_{m-1}=0.5 \\ f\left( 1-z_{m-1}, \alpha \right) . &{} 0.5<z_{m-1}<1 \end{array}\right. \end{array} \end{aligned}$$
(1)

where \(\alpha \in (0,0.5)\) is the control parameter and \(z_{0} \in (0,1)\) is initial condition. These two parameters serve as a secret key in the proposed watermarking scheme. Figure 1 shows the key sensitivity of PWLCM for a small change in initial condition. In Fig. 2, 5000 random numbers are generated for \(z_{0}=0.4999\) and \(\alpha = 0.2120\). From these figures, one can conclude that PWLCM is highly sensitive to initial conditions and all random numbers are different.

2.2 Substitution Box (S-Box)

S-Box is used to induce non-linearity in a cryptosystem. In order to develop an efficient and robust watermarking scheme, S-Box should be part of it. S-Box decreases correlation between original/plaintext data. High non-linearity in a cryptosystem is always of great interest. S-Box is fundamentally vector boolean function that maps n input bits to m output bits which means \(n \times m\) S-Box always produces m bits for a given number of n bits. In cryptographic theory, S-Box is a function of the form [2]:

$$\begin{aligned} {S-Box : GF(B)^{n} \rightarrow GF(B)^{m}.} \end{aligned}$$
(2)

Where \(B = {0, 1}\). As \(S-Boxes\) are nonlinear components, therefore, it determines security strength against differential and linear attacks.

3 The Proposed Scheme

Flowchart of the proposed image watermarking scheme is shown in Fig. 3. The presented scheme consists of two main steps: (1) dynamic S-Box generation using PWLCM chaotic map (2) embedding watermark in the original image. Detail steps of the proposed image watermarking scheme are as follows:

Fig. 3
figure 3

Flow chart of proposed scheme

Full size image

\(\mathbf {Step 1}\): Apply \(SHA-256\) on the original image P and generate the initial condition \(z_{0}\) for the PWLCM.

$$\begin{aligned} H_{bits} = SHA (P). \end{aligned}$$
(3)

\(\mathbf {Step 2}\): Convert 256 bit (64 byte hexadecimal) hash value to decimal format.

$$\begin{aligned} H_{decimal} = bi2de (H_{bits}). \end{aligned}$$
(4)

\(\mathbf {Step 3}\): Divide decimal hash value \(H_{decimal}\) by \(2^{48}\) to ensure that the value of initial condition \(z_{0}\) is between 0 and 1.

$$\begin{aligned} z_{0} = \frac{H_{decimal}}{2^{48}}. \end{aligned}$$
(5)

\(\mathbf {Step 4}\): Set maximum number of iterations i.e., 356 for PWLCM. To avoid the transient effect, the first 100 random numbers were ignored and the remaining 256 random numbers were considered. \(\mathbf {Step 5}\): Multiply random numbers vector z with \(10^{14}\) and apply modulus 256 operation.

$$\begin{aligned} R = Mod (z \times 10^{14}, 256). \end{aligned}$$
(6)

\(\mathbf {Step 6}\): Arrange decimal random numbers in \(16 \times 16\) matrix form and obtain dynamic S-Box.

$$\begin{aligned} \begin{pmatrix} S_{11} &{} S_{12} &{} \ldots &{} S_{1n} \\ S_{21} &{} S_{22} &{} \ldots &{} S_{2n} \\ \vdots &{} \vdots &{} \ddots &{} \vdots \\ S_{m1} &{} S_{m2} &{} \ldots &{} S_{mn} \end{pmatrix} \end{aligned}$$
(7)

\(\mathbf {Step 7}\): To get a substituted watermark \(W'\), substitute each pixel of the watermark with the value of dynamic S-Box. \(\mathbf {Step 8}\): Convert each decimal pixel of substituted watermark \(W'\) to binary 8 bits at index i and j.

$$\begin{aligned} W'_{bits} = decimal2binary (W'). \end{aligned}$$
(8)

\(\mathbf {Step 9}\): Split the 8 bit binary pixel values into 4 LSBs and 4 MSBs.

$$\begin{aligned}&LSBs = W'_{bit}(1:4), \nonumber \\&MSBs = W'_{bit}(5:8). \end{aligned}$$
(9)

\(\mathbf {Step 10}\): Apply LWT on the original image P.

$$\begin{aligned}{}[LL, LH, HL, HH] = lwt2(P). \end{aligned}$$
(10)

\(\mathbf {Step 11}\): Replace HL and HH blocks of the host image with the LSBs and MSBs of the substituted watermark.

$$\begin{aligned}{}[LL, LH, HL, HH] = [LL, LH, LSBs, MSBs]. \end{aligned}$$
(11)

\(\mathbf {Step 12}\): Apply inverse lifting wavelet transform on the new blocks of host image to get the final watermarked image W.

$$\begin{aligned} W = ilwt2([LL, LH, LSBs, MSBs]). \end{aligned}$$
(12)

To obtain the watermark and original image P from the watermarked image W, all the above steps are applied in reverse order. The step-wise flowchart of the watermark extraction algorithm is illustrated in Fig. 4. As the initial condition, \(z_{0}\) serve as a secret key in the proposed scheme, therefore, at the receiver end, the receiver must know this key. Due to this reason, the flow chart in Fig. 4 does not have key generation steps.

Fig. 4
figure 4

Flow chart of watermark extraction

Full size image
Fig. 5
figure 5

Visual results of proposed watermarking scheme for Baboon image a Original Baboon image. b Watermark imagem. c Substitute watermark image. d Histogram of original Baboon image. e Histogram of watermark image. f Histogram of substitute watermark image. g Watermarked Baboon image. h Extracted watermark image. i Recovered original Baboon image. j Histogram of watermarked Baboon image. k Histogram of extracted watermark image. l Histogram of recovered original Baboon image

Full size image
Fig. 6
figure 6

Visual results of proposed watermarking scheme for Pepper image a Original Pepper image. b Watermark image. c Substitute watermark image. d Histogram of original Pepper image. e Histogram of watermark image. f Histogram of substitute watermark image. g Watermarked Pepper image. h Extracted watermark image. i Recovered original Pepper image. j Histogram of watermarked Pepper image. k Histogram of extracted watermark image. l Histogram of recovered original Pepper image

Full size image

4 Simulation Results and Evaluation

Original Baboon image, watermark image, ciphered watermarked image, watermarked Baboon image, recovered watermark and recovered Baboon image are illustrated in Fig. 5a, b, c, g, h and i, respectively. Similarly, original Pepper image, watermark image, ciphered watermarked image, watermarked Pepper, recovered watermark and recovered Pepper image are shown in Fig. 6a, b, c, g, h and i, respectively. To prove the robustness and efficiency of the designed watermarking scheme, Structure Similarity Index (SSI), Structure Dissimilarity Index (SDI), Structure Content (SC), Mutual Information (MI), energy, entropy, correlation, histogram tests, and classical attacks analysis are carried out for the proposed scheme. All these parameters are discussed in the following subsections along with obtained results.

4.1 Structure Similarity Index (SSI)

SSI test is used to compute similarity between two variables or (in our case) two images. The results of SSI primarily reveals pixel inter dependency relationship. Mathematically, it is computed as [1]:

$$\begin{aligned} SSI (W,R) = \frac{(2\mu _{W}\mu _{R}+C1)(\sigma _{WR}+C2)}{(\mu _{W}^{2}+\mu _{R}^{2}+C1)(\sigma _{W}^{2}+\sigma _{R}^{2}+C2)} \end{aligned}$$
(13)

where W represent watermarked image and R represents recovered image, respectively. The variables \(\mu _{W}\) and \(\mu _{R}\) compute the average values of W and R, respectively. The parameters \(\sigma _{W}\) and \(\sigma _{R}\) calculate variance of W and R, respectively. Similarly, the parameters \(\sigma _{WR}\) quantifies the covariance of image W and R. Likewise, the parameters \(C1 = (K1L)^{2}\) and \(C2 = (K2L)^{2}\) are utilized to stabilize the division with weak denominator, where \(L = 2^{\#bitsperpixel} - 1\), \(K1 = 0.01\) and \(K2 = 0.03\). The resultant SSI is 1 if two images are identical, otherwise it is between 0 and 1. From Table 1, SSI value of the proposed image watermarking is near to the ideal value 1. Hence, SSI test confirms that the watermarked image is closer to the original image.

4.2 Structure Dissimilarity Index (SDI)

SDI computes dissimilarities between two images i.e., host and watermarked image. Mathematically it is calculated as [4]:

$$\begin{aligned} SDI (W,R) = \frac{1-SSI}{2} \end{aligned}$$
(14)

From Table 1, one can see that the SDI value of the proposed image watermarking scheme is considerably small, which again confirms that the watermarked image is closer to the original image.

4.3 Structure Contents (SC)

SC is the measure of information content in an image. Mathematically, it is written as the ratio of the structural information contents of the original and watermarked images. SC is calculated as [4]:

$$\begin{aligned} SC = \frac{\sum _{i=1}^{a}\sum _{j=1}^{b}[O(i,j)]^2}{\sum _{i=1}^{a}\sum _{j=1}^{b}[W(i,j)]^2} \end{aligned}$$
(15)

In the above equation, O(i; j) represents original image and W(i; j) represents the watermarked image at index i and j while a and b illustrates the number of rows and columns, respectively. From Table 1, one can see that the values of SC is close to 1 for the proposed watermarking scheme that confirms the robustness of the proposed scheme.

4.4 Mutual Information (MI)

The statistical dependence between two random variables is determined by MI test. MI can also be used to quantify robustness efficiency against various image degradation [4]. Mathematically, MI can be computed as follows [4]:

$$\begin{aligned} MI =\sum \sum p(O,W) \log ( \frac{p(O,W)}{p(O)p(W)}). \end{aligned}$$
(16)

where p(O, W) computes the joint probability distribution function of original and watermarked images while p(O) and p(W) computes the probability density functions in the individual images, respectively. The lower values of MI in Table 1 is a proof of good watermarking.

4.5 Energy

The disorder in texture can be computed via energy of an image. Energy of an image determines the information in an image in terms of the sum of squared values [4].

$$\begin{aligned} E = \sum (E(i,j)^2) \end{aligned}$$
(17)

where E(i, j) is pixel value at position i and j. Low energy means that the entries in Gray Level Co-occurrence Matrix (GLCM) are almost equal while higher energy means that some entries in the GLCM have higher magnitudes. A lower value of energy is required in image watermarking. The computed value in Table 2 proves that the proposed watermarking scheme has small value of energy.

Table 1 Experimentally computed values
Full size table

4.6 Entropy

Entropy test is used to calculate the degree of randomness of digital images. In watermarking, entropy is of great interest to approximate whether the watermarked image is similar to the original image or not. By assuming that an algorithm which output \(2^{8}\) symbols, its ideal entropy should be 8 bits. For a good watermarking scheme, entropy value of the watermarked image should be near to the original image entropy value. Mathematically entropy can be calculated as [17]:

$$\begin{aligned} H = \sum \left[ p(x_{i})\log _{2}\left( \frac{1}{p(x_{i})}\right) \right] \end{aligned}$$
(18)

In the above equation, \(p(x_{i})\) represents the probability of random variable x at \(i^{th}\) position. The calculated entropy for the proposed watermarking scheme is shown in Table 2. From Table 2, one can see that entropy values for watermarked and host images are approximately equal. Thus the proposed watermarking scheme can resist entropy attacks.

4.7 Correlation Analysis

Correlation coefficient determine similarity between variables. Mathematically, the correlation coefficient is written as [17]:

$$\begin{aligned}&Cov(O,W) = E[O-E(O)] \times [W-E(W)] \end{aligned}$$
(19)
$$\begin{aligned}&CC = \frac{Cov(O,W)}{\sqrt{D(O)D(W)}} \end{aligned}$$
(20)

where

$$\begin{aligned}&E(O) = \frac{1}{1000}\sum _{x=1}^{1000}(O(x))\\&D(O) = \frac{1}{1000}\sum _{x=1}^{1000}(O(x)-E(O))^2 \end{aligned}$$

The variable Cov(O, W) represents the covariance of adjacent pixels between original image O and watermarked image W, respectively. Correlation coefficient values in horizontal, vertical and diagonal directions are highlighted in Table 2. From Table 2, one can see that the correlation coefficient values in all direction are almost same which confirms the similarity between watermarked and original images.

Table 2 Experimentally computed values
Full size table

4.8 Histogram Analysis

The histogram shows the pixel distribution of a digital image. From Figs. 5d, e, f, j, k and l and 6d, e, f, j, k and l, it is evident that the histogram of the watermarked image is similar to the original image histogram. Furthermore, the histograms of the recovered watermark and recovered plaintext original images are also exactly similar to the original watermark and original plaintext images histograms. Hence, the histogram plot confirms that an intruder cannot detect any differences between original and watermarked images.

4.9 Classical Attacks Analysis

SHA-256 and plaintext image are utilized to produce the initial values of PWLCM chaotic maps \((z_{0})\). This makes the proposed watermarking dependable on the plaintext image. For a single pixel change in the original plaintext image, the proposed scheme will produce a totally different random number. Therefore, during extraction and inverse substitution of the watermark image, instead of generating the original watermark image, the algorithm will produce a totally different cipher watermark. As a result, the proposed watermarking scheme can resist ciphertext-only, chosen-plaintext and known-plaintext attack.

5 Conclusion

In this paper, a dynamic S-Box based efficient watermarking scheme is presented and statistically evaluated. The dynamic S-Box and PWLCM strengthen the security of the proposed scheme due to the dynamic nature. The initial condition for PWLCM was generated from the \(SHA-256\). The watermark is substituted with the values of dynamic S-Box. Due to the dynamic nature of S-Box, it is hard for the eavesdroppers to launch an attack on the proposed scheme. Lastly, LWT is applied on the host image and HL and HH blocks of host image were replaced with Least Significant Bits (LSBs) and Most Significant Bits (MSBs) of the substituted watermark. The efficiency and robustness of the presented watermarking scheme are tested against numerous security parameters which confirmed that the scheme is highly secure. Furthermore, at the receiver’s end, the proposed scheme can recover the original image without loss of any sensitive information. In future, we intend to optimize the proposed scheme by targeting the colored images rather than only gray-scale ones and we will also evaluate the scheme under numerous types of attacks.