Part I Sciencific achievements
3.2 Information Security Solutions
3.2.2 Confident System Architecture
Chatterjee et al. (2015) identified different network entities for their proposed confident architecture [R 17], which is given in Figure 20.
These entities are:
• The User: Customers who want to use cloud infrastructure.
• Cloud Service Provider-1 (CSP-1): Cloud Infrastructure where the data will be stored in the form images.
• Cloud Service Provider-2 (CSP-2): For storing the encryption and decryption techniques - mechanism for hiding data into images and retrieving data from images.
• Cloud Service Provider 3 (CSP-3): Interact with CSP-1 and CSP-2 - all calculations for the user are done here
Figure 20 The architecture of our proposed model [R 17]
3.2.2.1 Security Model
Instead of storing data into a file, the data are hidden and stored into images. This is done by a steganography process [R 183]. This is the new paradigm of security through obscurity.
Therefore, steganography is the new paradigm of security through Obscurity. As example, in [R 17], an entire file is divided into multiple parts, and each part is stored into a corresponding image. The number of divisions in which the file is divided depends on file size and image size.
The proposed security model for storing data is in Figure 21 and the proposed model for retrieving data is given in Figure 22. In CSP-3 are done the computations for users.
Figure 21 Computational model for storing data [R 17] Figure 22 Computational model for Retrieving Data [R 17]
All temporary file must be deleted when the user log out from the system. The processes of storing data and retrieving data are dealing with:
• The Image Database - contains a set of images of different sizes stored in CSP- 1. A set of images will be sent to CSP-3 when a user wants to store data into cloud.
• The File Database - is a file and this file holds the address of images where we will store our data; does not contain the actual information
• Embedding Data into Images process– this process generates files which are to be stored in cloud data storage. It counts the total no of character presented in the file and find the frequency of the occurrences of each character and codifies the original characters by Huffman codes. After that steganography is applied to both frequency of characters and the codified data.
3.2.2.2 Analysis of Wavelet, Ridgelet, Curvelet and Bandelet transforms for QR code based Image Steganography
In [R 87] we propose a data hiding based image steganography method that uses to perform image steganography four transforms from frequency domain. The experimental results demonstrate that the proposed method provides better stego- image quality and increases the embedding capacity. It maintains for the stego-image quality the PSNR value of above 47 dB without affecting the retrieved secret message.
In the spatial domain, the information is hidden directly into the least significant bit (LSB) of the cover image without any modification. In [R 41] [R 11] [R 215] are presented methods and security problems related with the hiding operation in the spatial domain and in [R 133] is proposed an iterative method of palette based image steganography using spatial domain. For increasing security which is the major issue in the spatial domain, some channel selection criterion has been proposed [R 215]. In the frequency domain, the information is hidden in the transform coefficients. It transforms the spatial domain cover images into frequency domain cover images using Discrete fourier transform (DFT), Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), ridgelet, curvelet and bandelet transform [R 87].
Reversible data hiding is also a research area for sensitive applications like medical, military and wireless; it retrieves the secret message and original cover image from the stego image with the same secret key on both the transmitter and receiver side. In the spatial domain, in [R 11] is proposed a reversible data hiding in the digital content using difference expansion and in [R 214] is proposed a high capacity reversible data hiding scheme based on generalized prediction-error expansion and adaptive embedding strategy. In the frequency domain, a lossless and reversible data hiding scheme in DCT-based compressed images have been proposed. The information is embedded in DCT coefficients of JPEG images [R 14] [R 87].
To achieve the high embedding capacity and good quality of stego image in [R 15] is proposed to use Haar DWT based on reversible data hiding. [R 120] propose the use of the curvelet transform for data hiding based on amplitude. The bandelet transform for data hiding have been proposed in [R 188]. The data hiding process changes the statistical properties of the image, which leads to steganalyst attempts to detect the statistical traces called statistical steganalysis. To resist the statistical steganalysis, reversible histogram transformation function based on LSB steganography technique has been proposed in [R 131] [R 87].
Image steganography using frequency domain method is found to be more efficient than the spatial method due to its high data security [R 87].
So, in [R 87] we briefly presented the five frequency domain techniques: (1) QR code (2) Integer discrete wavelet transformation (3) Ridgelet transformation (4) Curvelet transformation (5) Bandelet transformation, which are used for our study.
The QR code is used by our proposed method as a secret message to increase the security and embedding capacity of the steganography [R 87].
Integer Discrete Wavelet Transform (IDWT) is an important technique for transforming a cover image from spatial domain into frequency domain. The operation is performed through cohen-daubechies-feauveau wavelets in the three lifting steps for forward transform (splitting, prediction, update, and three lifting steps for inverse transform (inverse update, inverse prediction and merging). IDWT decomposes the cover image into approximation co-efficient (LL) which is low frequency band and detail co-efficient (LH, HL, HH) which is high frequency band. The high frequency band (LH, HL, HH) of IDWT hides the QR coded secret message [R 87].
The Finite Ridgelet Transform (FRIT) was proposed based on the Finite Radon Transform (FRAT) [R 87].
The curvelet transform was proposed by Candes and Donoho, to overcome the shortcomings of wavelet transform. It is a multiscale directional transform that represents the object with edges and it requires only fewer amounts of coefficients to represent the edges. There are two transformations related to the Fast Discrete Curvelet Transform (FDCTs) [R 87]:
1. Using Unequally-Spaced Fast Fourier Transform (USFFT)
2. The wrapping of specially selected fourier samples is the basis for the second one.
Our proposed method uses the FDCTs via USFFT [R 87].
Bandelet transform is mainly used to represent efficiently the edges and texture of the image. It takes the sharp transitions in the image as an advantage. The bandelet transform overcomes the wavelets’ high dimension problem[R 87].
Proposed Methodology
To perform image steganography using frequency domain, we have selected: Integer Discrete Wavelet Transform (IDWT), Finite Ridgelet Transform (FRIT), curvelet transform and Bandelet transform.
Our embedding technique is based on the bit-plane compression method. adaptive histogram modification is carried out as a pre-processing of images in order to prevent overruns occur during the embedded process. The changes in the selected bit-plane of high frequency band are indicated by the overflow that happens when the grayscale value of the QR pixel code exceeds one of the two boundaries: the lower bound (0) or the upper bound (255). The QR coded secret message is embedded into the cover image by the embedding phase. The embedding algorithm is common for all the four transforms [R 87].
The concept of the proposed method is that, QR code can be embedded in the one or more bit-planes of transform’s high frequency sub-bands. The space to hide the QR code is generated by compressing the bits in a bit-plane [R 87].
For instance, let us see the eight-bit binary data 01101101, from the left, the bit binary data 1 represents LSB and the binary data 0 at the eighth bit plane represents MSB.
Let the bit-plane be represented by the notation k. QR code can be embedded in any of the selected bits-plane going from first bit-plane to the eighth bit-plane [R 87]:
In the frequency domain, the transform (IDWT, FRIT, FDCT via USFFT and bandelet transform) decomposes the cover image into low and high frequency sub-bands and the QR code can then be embedded in the transform’s high frequency sub-bands. QR code can be embedded in any of the selected bit-plane with the secret key. The secret key is used to enhance the information security. The secret key is x-or-ed with the QR code. The bits in the selected bit-plane can be compressed using arithmetic coding to leave a space to hide the QR code. The extraction phase is an inverse process of the embedding phase. The extraction phase extracts the cover image and the QR code with the same secret key. The image quality varies depending on the nature of the transform [R 87].
The algorithm for the embedding phase is given below [R 87]:
1. Step 1: Adaptive histogram modification is used as a preprocessing.
2. Step 2: Decompose the cover image by IDWT, FRIT, FDCT via USFFT and Bandelet transform separately.
3. Step 3: Select the bit-plane (k =1, 2, 3, 4, 5, 6 and 7) of transform’s high frequency sub-bands.
4. Step 4: Compress the data in selected k using arithmetic encoding to leave a space to hide QR code.
5. Step 6: Convert the secret data into a QR code and the secret key is embedded into a QR code for information security.
6. Step 7: Embed the QR code in the space left.
7. Step 8: Compute Inverse wavelet, ridgelet, curvelet and bandelet to get the stego image.
In our work, the five pre-processed cover images are tested for k =1, 2, 3, 4, 5, 6 & 7 to find the k value which has the best stego image quality. Arithmetic coding is used to compress the bit-plane to leave a space for hiding the QR code. It replaces an input symbol with some specific code [R 87].
Experimental Results and Discussions
The evaluating the performance of the proposed method many simulations experiments were done. Three commonly used gray-level images: “Baboon”,
“Barbara”, “Lena” and two medical gray-level images: “Eye”, “Skull” (totally five gray level images) were used. For better result, the cover image was pre-processed by adaptive histogram equalization. The pixels of each pre-processed gray-level images are 512*512 which is used as a cover images are shown in Figure 23.
MATLAB R.2010 software was used.
Peak Signal-to-Noise Ratio (PSNR) and embedding capacity were the performance measures before an extraction, and after extraction, the performance measures used were: Tamper Assessment Factor (TAF) and Normalized Absolute Error (NAE). TAF is used to determine the credibility of image authentication, which is measured between the QR code and retrieved QR code. NAE is measured between the cover image and the restored cover image [R 87].
The relations for these performance measures are using the following notations: C(i,j) is the cover image, S(i,j) is the stego image, Q(i,j) is the QR coded secret message, Q’(i,j) is the retrieved QR coded secret message and C’(i,j) is the restored cover image, then:
𝑀𝑆𝐸 =&∙(% &32% (12% 𝐶 𝑖, 𝑗 − 𝑆(𝑖, 𝑗) 0 (3.1)
𝑃𝑆𝑁𝑅 = 10𝑙𝑜𝑔%< 0==?@A> (3.2)
𝑇𝐴𝐹 =&(% &32% (12% 𝑄(𝑖, 𝑗)⨁𝑄′(𝑖, 𝑗) (3.3) 𝑁𝐴𝐸 = OPMN KLMNH 3,1 IHJ(3,1)
H(3,1) KLMN
OPMN (3.4)
For all our experiments The QR code is hidden in the third, fourth, fifth, sixth and seventh bit-plane (k = 3, 4, 5, 6 and 7) of the high frequency sub-bands coefficients of IDWT (LH, HL and HH) of “Lena” image. By testing the “Lena” image for all the k values, the best stego-image quality is obtained for k =7. Hence, for all experiments, the selected bit-plane taken is for k =7 and applied to all the pre-processed cover images [R 87].
Results of IDWT
The proposed algorithm is applied to the IDWT based steganography. The resultant stego-images corresponding to the cover images are shown in Figure 23. Table 8,
lists the PSNRs and the embedding capacity of the stego-images for all the k values [R 87].
From Table 8, it is evident that, in the commonly used images, the PSNR value is high for “Barbara” stego-image, but the embedding capacity is high for “Baboon” stego- image. In the medical images both PSNR value and embedding capacity is high for
“Skull” stego-image when k= 7. Hence, the result varies according to the nature of the images, when performed in IDWT [R 87].
Results Of FRIT
The proposed algorithm is applied to the FRIT based steganography. The resultant stego-images corresponding to the cover images are shown in Figure 24 Table 9, lists the PSNRs and embedding capacity of five stego images when k =1, 2, 3,4, 5, 6 and 7 [R 87].
From Table 9, it is evident that, in the commonly used images, the PSNR value is high for “Lena” stego-image, but the embedding capacity is high for “Baboon” stego-image.
In medical images, the PSNR value is high for “Skull” stego-image. Hence, the result also varies according to the nature of the images, when performed in FRIT [R 87].
Results Of FDCT Via USFFT
In the third experiment, the proposed algorithm is applied to the FDCT via USFFT based steganography image. The resultant stego-images corresponding to the cover images are shown in Figure 25. Table 10, lists the PSNRs and embedding capacity of stego images for all the k values. FDCT provides a better result compared to the other transforms [R 87].
From Table 10, it is found that, in the commonly used images, the PSNR value is high for “Barbara” stego-image, but the embedding capacity is high for “Baboon” stego- image. In medical images, the PSNR value is high for “Eye” stego-image. Hence, a result also varies according to the nature of images, when performed in FDCT [R 87].
(a)k=7, 50.02 dB (b)k=7, 50.5 dB (c)k=7, 50.3 dB (d) k=7, 49.1 dB (e) k=7, 50.7 dB Figure 23 Stego image using IDWT (a) Baboon, (b) Barbara, (c) Lena, (d) Eye and (e) Skull [R 87]
k 3
(PSNR & EC)
4 (PSNR & EC)
5 (PSNR & EC)
6 (PSNR & EC)
7 (PSNR & EC)
Baboon 24.6 220839 29.7 260523 35.3 277980 40.9 284616 46.0 286613 Barbara 24.7 164844 29.5 202127 35.3 234438 40.8 260242 46.1 282077 Lena 24.9 138000 29.6 187032 35.2 257531 40.9 281449 46.1 286276 Eye 22.8 159067 27.6 169092 33.4 164131 39.3 245276 44.6 282148 Skull 24.8 149067 30.9 149048 35.1 142012 40.7 220305 46.2 279085 Table 8 PSNR value (dB) & embedding capacity for stego image of IDWT when k =3, 4, 5, 6, 7 [R 87]
(a) k=7, 50.4 dB (b) k=7, 50.4 dB (c) k=7, 50.6 dB (d) k=7, 48.7 dB (e) k=7, 50.4 dB Figure 24 Stego image using FRIT of (a) Baboon, (b) Barbara, (c) Lena, (d) Eye and (e) Skull [R 87]
k 3 (PSNR & EC)
4 (PSNR & EC)
5 (PSNR & EC)
6 (PSNR & EC)
7 (PSNR & EC) Baboon 24.3 149220 29.5 155985 34.9 157439 40.6 157690 45.8 157752 Barbara 24.2 138012 29.3 153533 34.8 156801 40.5 157527 45.9 157728 Lena 24.1 135682 29.1 151383 34.9 156421 40.9 157456 46.1 157697 Eye 22.6 105691 26.8 133529 33.2 151229 39.1 156448 44.4 157453 Skull 24.1 133919 29.2 149830 34.7 156071 40.6 157410 45.9 157683
Table 9 PSNR value (dB) & embedding capacity (bits) for stego image of FRIT when k =3, 4, 5, 6, 7 [R 87]
(a) k=7, 82.8 dB (b) k=7, 86.7 dB (c) k=7, 85.4 dB (d) k=7, 93.7 dB (e) k=7, 90.6 dB Figure 25 Stego image using FDCT of (a) Baboon, (b) Barbara, (c) Lena, (d) Eye and (e) Skull [R 87]
k 3
(PSNR &
EC)
4 (PSNR &
EC)
5 (PSNR &
EC)
6 (PSNR &
EC)
7 (PSNR & EC)
Baboon 82.8 82.8 91203 82.8 91209 82.8 91208 82.8 91209 82.8 91209 Barbara 86.7 86.7 91208 86.7 91206 86.7 91207 86.7 91206 86.7 91208 Lena 85.4 85.4 91204 85.4 91206 85.4 91206 85.4 91209 85.4 91209 Eye 93.7 93.7 91208 93.7 91203 93.7 91207 93.7 91207 93.7 91209 Skull 90.6 90.6 91192 90.6 91205 90.6 91209 90.6 91208 90.6 91207 Table 10 PSNR value (dB) & embedding capacity (bits) for stego image of FDCT when k=3, 4, 5, 6 & 7 [R 87]
(a) k=7, 47.8 dB (b) k=7, 48.1 dB (c) k=7, 47.9 dB (d) k=7, 47.4 dB (e) k=7, 48.8 dB Figure 26 Stego image using bandelet transform of (a) Baboon, (b) Barbara, (c) Lena, (d) Eye and (e)
Skull [R 87]
k 3
(PSNR & EC) 4
(PSNR & EC) 5
(PSNR & EC) 6
(PSNR & EC) 7
(PSNR & EC) Baboon 21.8 310082 26.5 339076 31.9 347613 37.4 350860 42.9 351102 Barbara 21.8 255146 26.6 291495 31.9 318083 37.5 337516 43.0 346424 Lena 21.8 214182 26.4 271040 31.8 321103 37.4 344101 42.9 348321 Eye 22.2 198313 26.5 211691 31.1 270798 36.6 316730 42.6 338329 Skull 22.8 204271 27.6 241055 32.7 294760 38.4 328546 43.9 343150
Table 11 PSNR value (dB) & embedding capacity (bits) for stego image of bendlet when k=3, 4, 5, 6 & 7 [R 87]
Results Of Bandelet Transform
In the fourth experiment, the proposed algorithm is applied to the bandelet transform based steganography. The resultant stego-images corresponding to the cover images are shown in Figure 26 Table 3, list the PSNRs and embedding capacity of the stego- images for all the k values.
From Table 3, it is found that, PSNR value is high for “Barbara” stego-image, but the embedding capacity is high for “Baboon” stego-image. Hence, the result also varies according to the nature of images, when performed in bandelet transform [R 87].
Comparative Analysis For The Proposed Approaches
In Figure 27 and Figure 28 is given the comparative analysis performed for the proposed method are given.The PSNR value is high for FDCT, low for bandelet transform (Figure 27) and embedding capacity is high for bandelet transform and low for FDCT (Figure 28.). For medical applications, the steganography is used to hide the patient medical report with his/her scanned image and can be send to clinicians residing in any corner of the globe for diagnosis.
Figure 27 Comparison of PSNR value of the four transforms [R 87]
Figure 28 Comparison of embedding capacity of the four transforms [R 87]
When the patient medical report capacity is low, FRIT based steganography is preferable as it gives better quality of stego image when the embedding capacity is low. For secured communication applications like military domain, FDCT based steganography is preferable because it gives good quality of stego image. For academic applications like transforming more data within the local network or within the campus, bandelet transform based steganography is preferable as the time consumption is low and acceptable PSNRs value is provided. To embed the high amount of information and to attain quality of stego image IDWT based steganography is preferable, it also provides better results and is less time consuming to embed the information [R 87].
The comparison of the proposed technique with other works shows that the stego- image quality of the proposed method is significantly higher than those of the other works mentioned in literature survey. The embedding capacity is also increased in our proposed method compared with the other work [R 87].
3.2.2.3 Application of Genetic Algorithm and Particle Swarm Optimization techniques for Improved image steganography systems
Though the frequency domain techniques are preferred for image steganography applications, there still are significant drawbacks associated with these techniques.
Thus, in transform based approaches, the transform coefficients that embed in random manner the secret data may not be optimal in terms of the stego image quality and embedding capacity.
So, in [R 89] is explored, in the context of determining the optimal coefficients in these transforms, the application of Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). The proposed methodology increases the fidelity of the stego image and embedding capacity and also provides more security, by combining frequency domain transforms such as Bandelet Transform (BT) and Finite Ridgelet Transform (FRIT) combination with GA and PSO. BT and FRIT were chosen to yield the high embedding capacity and GA and PSO were used to find the most significant coefficients for better information hiding. It is a novel attempt to enhance the efficiency of the steganography system.
The experimental results of the proposed approaches has been compared with recent works [R 89].
Particle Swarm Optimization(PSO)
Particle Swarm Optimization (PSO) is a population based optimization techniques, which has been developed by Kennedy and Eberhats in 1995. The potential solution is represented by each individual. Each particle’s position is altered according to its neighbors and with its own practical experience. In each iteration, the predetermined particles correspondingly produce fitness value from the fitness function and also have velocity to direct the movement of the particle. Each particle in a population keeps track of its best solution (fitness) in the search space which has achieved so far by that particle. This fitness value is called pbest (personal best). PSO keep track of another best solution that is obtained so far by any particle in the neighbor- hood of that particle.
This is known as gbest (global best) [R 89].
A detailed algorithm is given in [R 4].
The Proposed Methodology
The proposed methodology is based on two phases: the embedding phase and the extraction phase. In Figure 29 is depicted by the block diagram of embedding phase [R 89].
Embedding Phase
The first step of embedding phase is to read the cover image 𝐴(𝑥, 𝑦). The cover image is decomposed using specific transforms (BT & FRIT). Then, the most significant coefficients are selected using GA and PSO. Embedding the secret data in the most significant coefficients, that will increase the fidelity for the stego image.
Next step is to read the secret data B’(x,y).
The secret data B’(x,y) is hidden in the most significant coefficients [R 89].
Extraction Phase
The extraction phase extracts the
embedded secret data and cover image separately. Decompose the stego image using specific transform (BT and FRIT). Then use the positions of most significant coefficients to determine the extraction key. The extraction key posses the position of most significant coefficients [R 89].
GA base data embedding
The procedure of GA for finding the most significant coefficients in BT and FRIT is given below [R 89].
Step 1. Parameter Representation; these are given in [R 89].
Step2. Fitness Function
To enhance the quality of stego image, Peak Signal-to-Noise ratio (PSNR) equation is taken as a fitness function. GA and PSO search for the chromosomes with highest fitness value from the fitness function. PSNR is measured between cover image and stego image. If PSNR value is high, the fidelity of stego image is also high [R 89].
MSE =V∙W% Vb2% Wa2%[A x, y − Stego x, y ]0 (3.5)
Figure 29 Block diagram of Data embedding [R 89]
