RIWT and QR factorization based hybrid robust image steganography using block selection algorithm for IoT devices

Abstract

The development of the Internet of Things (IoT) can be attributed to the sudden rise in miniature electronic devices, as well as their computing power and ability to make interconnections. These devices exchange large volumes of confidential information from diverse locations. Similar to the Internet, the IoT has also encountered various issues with information security. Due to limited computing and energy resources in the field of IoT, it is necessary to develop a scheme to ensure feasible and more effective concealment and security properties. This paper proposes a unique methodology that captures an image using IoT sensors, which are subjected to lighter cryptographic operations for conversion into a cipher image, and is then sent to a home server. At the home server, a combined cryptography and steganography approach is employed to conceal the cipher image in a cover image, camouflaging the presence of the secret image, which is then sent to the IoT-Cloud server for storage. During the embedding process, QR decomposition is performed on the RIWT transformed secret image and RIWT - DCT transformed cover image. Modification performed on the R matrix of QR decomposition does not affect the structural properties of the cover image. A block selection algorithm is used to select optimal blocks with high contrast areas to embed the secret image. The experimental results indicate that our scheme enhances imperceptibility, robustness, and resistance to steganalysis attacks.

Keywords

Image steganography RIWT DCT security in cloud block selection algorithm QR decomposition IoT

1 Introduction

The IoT provides services to its consumer using an interconnected infrastructure that merges the digital and physical world. IoT enhances communication between users, devices, and applications that are scattered in various locations across the world. The public is increasingly exposed to a diverse range and number of computing devices with different capabilities and sizes that facilitate interactions amongst the IoT. A larger number of personalized gadgets, such as smartphones, laptops, tablets, and other such smart devices, are being linked to the Internet in order to access various services for everyday purposes in this new digital environment [1]. The Internet of Service is the primary component of the Internet, wherein the availability of an enormous number of services are externally available across the Internet, improving productivity [2]. The IoT has inspired service provides to present applications on the network convergence for coherent services. Statistics show that more consumers have started using IoT service for their day-to-day activites [3]. IoT applications are versatile, ranging in use from agriculture, home, medicine, defense, science, research, logistics, environment, etc. Security is a particular concern, considering the numerous applications of IoT, making it a critical issue for the growth of this technology [4]. IoT insecurity was listed among the world’s top five security threats in 2015, prompting research community to collaboratively apply research skills to provide increased security and to tap the exponential benefits of this technology in a global arena [5].

The literature discusses many methods to provide security for confidential information shared between users that are separated geographically separated. These methods can be classified into two types: cryptography and steganography. Cryptography provides security, but the existence of secret in cipher can be suspected, while steganography provides security but does not reveal the existence of secret in stego [6].

More specifically, cryptography encrypts plain text into a ciphertext based on an encryption algorithm as well as an automatic and randomized key, completely outwitting the interceptor about the existence of the original confidential data, so only authorized persons can decipher and receive messages. Steganography completely hides the existence of the secret by embedding secret data concealed in a cover image in such a manner that it becomes almost impossible to intercept confidential data and hack it [7].

To utilize the advantages of both security techniques, a security model is designed for IoT that integrates both steganography and cryptography to efficiently defend against malicious threats [8]. Security can be provided in two phases, as shown in Fig. 1. In phase-1, IoT sensors capture a medical image and are subjected to a lighter encryption operation using the lightweight chaotic map as it possesses limitations like battery power and memory capacity. In phase-2, a home server conceals encrypted images in a cover image using steganography techniques. This scheme provides security on two levels; one at the IoT sensor end and the second at the home server end. This security system secures data transmission in a smart environment by combining encryption and steganography and is safe enough to be applied in the field of IoT [9].

Fig.1

The architecture of steganographic scheme for IoT-Cloud.

The cover selection approach in steganography is a novel approach that selects an optimal cover from a pool of images from a standard database to conceal the secret information. The cover selection makes the stego and cover image look alike so that steganalysis is more difficult [10].

Our article presents a robust hybrid steganographic scheme for IoT using RIWT, DCT, and QR. An IoT sensor captures a medical image and encrypts it using a lightweight chaotic map algorithm and then sends it to the IoT-Cloud server. The cover image is first decomposed into 16×16 pixel sized non-overlapping blocks. Using the block selection algorithm, the best blocks in the cover are selected, and from these, the embedding map is obtained. According to the value of this embedding map, the chosen block is subjected to DCT and RIWT transformations independently. The secret image is also decomposed into 8×8 pixel sized non-overlapping blocks. The decomposed secret image block is concealed in the selected decomposed cover image block. QR decomposition is applied to both the secret and cover image blocks. The R matrix of the cover image block is modified to embed the R matrix of the secret image block. Modification performed on the R component of QR decomposition has less of an overall effect on the cover image. The proposed robust steganographic scheme is an enhancement of approaches available in the literature, as follows:

The cover image is decomposed into blocks of smaller size. Using a block selection algorithm, the optimum block is chosen for embedding, enhancing imperceptibility and protecting the stego image against steganalysis.

RIWT, DCT use considerably enhances the imperceptibility of the produced stego image, as embedding performed on the less significant coefficient of DCT transform provides less distortion of the stego image.

Embedding is performed on the R matrix of QR decomposition, which produces less distortion in the cover image, enhancing robustness.

The remainder of the article is presented as follows: Section 2 lists similar contributions available in the literature. Section 3 describes Block selection algorithm; Section 4 describes our proposed scheme. Section 5 discusses the results of our proposed work and compares our findings with similar schemes. Finally, Section 6 provides a conclusion and considers ideas for future research.

2 Literature survey

Many healthcare applications that apply wireless medical sensor network (WMSN) for IoT environment were surveyed. Various healthcare system security issues, methods and security techniques, especially hybrid security techniques are also discussed in [11]. To enhance medical image security, verification systems must ensure medical image integrity. This is done in two phases, the protection phase, and verification phase. In the protection phase, the cover image is decomposed using DWT transform. In HH sub-band, the secret image is embedded to produce a stego image. The extraction phase transforms the stego image based on DWT, and the secret image is extracted [12]. An image encryption technique uses shifted image blocks and an AES algorithm. A shifted algorithm is deployed to disintegrate an image into blocks of some fixed size. Then these blocks are shifted by row and then by column. Later, these shuffled blocks are encrypted using an AES encryption algorithm to produce a cipher image [13]. Various security vulnerability and risk factors are present in the mobile medical platform. Based on risk factors, mobile medical apps are classified by medical information, remote monitoring, diagnostic support, treatment support, etc. [14]. An image steganography scheme using an Inverted LSB technique for transmitting face images in the IoT environment is designed. A face image is captured through an IoT sensor camera, and due to its low computational power, lightweight encryption is done here and then transmitted to a home server, where a heavier protection method is initiated, and then sent to IoT-Cloud [15]. Image encryption techniques are applied to protect medical images. This is done not only to preserve confidentiality but also to maintain the integrity, availability, and authenticity of medical images. An AES algorithm is used to preserve confidentiality. The features of an ear image are extracted and then embedded into a cover image to produce a stego image, which can be transmitted through the internet with no fear of being disclosed to an unauthorized user [16]. Quality of stego image degrades while embedding medical images into a cover image. First, a medical secret is encrypted using an RC4 algorithm, and a selected cover image is transformed using DFT. Image degradation is more apparent if a cipher is embedded into low-frequency bands, and less apparent if embedded into upper-frequency bands [17].

Using a Remote Patient Health Monitoring System (PPHM), a patient’s health can be examined by IoT sensors kept inside their home, providing more reliable results as patients are more comfortable in their own homes. At the home server end, the patient’s health-related vital information is embedded into a cover image to produce a stego image and then transmitted to the doctor for further processing [18]. Security threats are present both from within the IoT environment and from the public network. IoT sensors do not provide much computational power so that some lighter security mechanisms can be implemented, but the home server can provide stronger cryptographic and steganographic function. By employing security measures in these two places, security threats can be minimized [19]. Privacy leakage vulnerability is studied in remote home monitoring systems. There are five vulnerable points: at the surveillance camera, during transmission of captured information, at the home server, at the local storage server, at the global storage server, and at the coordinator servers. In such scenarios, efficient secret sharing techniques are required to secure the transaction [20]. Public key encryption and digital signatures can also be used to minimize the above scenario. Both techniques are employed to provide confidentiality and mutual authentication [21]. It is possible that encrypted patient data may be lost or stolen during the transaction. Attackers could try to modify the contents, resulting in data distortion or content unavailability for a period. Timestamps can be used to prevent such security threat [22]. Numerous matrix decomposition techniques are available, including singular value decomposition (SVD), QR decomposition, LU decomposition, and Schur decomposition. Among these, SVD is globally used for image steganography. SVD can be used in two approaches; the pure SVD method conceals a cover message directly into singular values in a cover image, and the hybrid SVD, both cover images, and secret images can be transformed and then SVD is employed [23]. In the literature, more steganographic approaches are based on SVD. Due to this, usage of SVD in the embedding and extraction phase, imperceptibility, capacity and robustness measures are improved [24]. Despite the many advantages of using SVD, there are also some disadvantages. During steganalysis of the performance analysis measure, the stego image is always assessed to classify it as a stego image or cover image, i.e., to measure the rate of true positives and false positives. However, during steganalysis of SVD-based approaches, false positives are frequent as the stego image is often misclassified as the cover [25]. To overcome this, other matrix decomposition approaches, like QR decomposition, are used in image steganography. The results are then compared with the results of schemes that use SVD. The analysis shows that resistance to steganalysis increases considerably in QR based schemes [26]. Contourlet transform is used on both the cover image and secret image as HVS does not clearly distinguish the edge portion of the image. As the low frequency of transformed coefficients represents the highest energy, this band is preferred for hiding secrets. Low-frequency bands in both the secret and cover image are decomposed using QR decomposition methods and then embedding is performed. Imperceptible and robust values perform better against many attacks using image processing techniques, like salt and pepper noise, Gaussian noise, scaling and compression, [27]. To provide security, a secret image can be encrypted using an image encryption algorithm. However, today, instead of using traditional image encryption algorithms, a chaotic image encryption algorithm is used. Unfortunately, due to heavy operations, such as performing specific steps over many rounds, these algorithms cannot be used for resource constraint devices. To address this issue, a lightweight chaotic image encryption algorithm is designed [28].

In the past, embedding position preference was treated as the foremost design criteria for designing foolproof steganographic approaches. The cover image selected to hide a secret image plays a prominent part in designing a new steganographic approach, because if it is done wisely, steganalyzer may not classify properly between stego image and cover image. Watson’s metric, JPEG quality factor, mean square error (MSE) and prediction error are some of the parameters used to find the optimal cover image for embedding [29]. The texture is a fundamental property of an image. In the various available approaches for texture analysis, statistical methods are widely used. In this method, local features are tested at every image pixel and a list of statistics is obtained through consideration of the spatial interaction between image pixels and their neighbors [30]. Today, transform domain steganography uses IWT transform as it introduces less error in forward and inverse transformations.

Similarly, RDWT provides robustness to steganography schemes. To utilize the advantages of both schemes, RIWT transform techniques are used because RIWT integrates both IWT and RDWT [31]. This property allows DCT to manifest most of the signal data in a transformed domain that is concentrated at a small portion at a low-frequency and maintains a vast area in the high-frequency region of DCT domain with a less important co-efficient. This huge area acts as an embedding area, since replacement of this unimportant co-efficient will not impact the quality of the stego image, making it possible to reach a high embedding capacity if above 20 bpp while the stego quality is maintained at an average level of 30 dB [32].

3 Block selection algorithm

In [10], the cover selection algorithm for selecting the best cover from a pool of covers in an image database is explained. This algorithm is slightly modified to suit the requirements of our proposed algorithm. The following steps are applied for every cover image block to obtain a co-occurrence matrix by using the relation between every pixel,

Step 1: Compute co-occurrence matrix C (x, y) by calculating how often pixel value×occurs horizontally adjacent to a pixel with value y. The number of unique pixel value determines the size of the co-occurrence matrix.

Step 2: Every element in the co-occurrence matrix gives the probability of pixel pairs as shown in Fig. 2.

Step 3: Contrast is a feature in which the amount of difference between neighbors is computed as follows, $\sum_{i, j} | i - j |^{2} P (i, j)$

Step 4: Correlation is another feature in which the similarity between neighbors is computed as follows, $\sum_{i, j} \frac{(i - μ_{i}) P (i, j)}{σ i σ j}$

Step 5: Based on the size of the secret image block, the number of cover image blocks necessary to embed is computed.

Step 6: A threshold value is set for both contrast and correlation values.

Step 7: Values of contrast and correlation for each block are compared with the threshold value to determine it satisfies the selection criteria.

Step 8: The number of blocks chosen in step 6 is compared against step 4. If there are not enough, the threshold value is adjusted so that the required blocks are chosen.

Step 9: The embedding map for a chosen cover is constructed, it contains 1 for a block if it satisfies the block selection criteria, otherwise 0.

Fig.2

Co-occurrence matrix.

4 Proposed methodologies

In this section, a Hybrid Robust Image Steganography scheme is proposed for IoT devices. The images that are captured through IoT sensors are first subjected to lighter cryptographic operations, such as a lightweight chaotic map operation, to convert them into ciphers. Cipher images are transmitted to the home server where they are segmented into blocks of 8×8 pixels. An appropriate cover image is chosen to hide our confidential images. The chosen cover image is segmented into blocks of 16×16 pixels. A co-occurrence matrix is calculated for each block to determine whether the blocks can be used for embedding. According to the co-occurrence matrix for each block, an embedding map is created for a cover image. Using an embedding map, a cover image block is transformed by DCT and then by RIWT. QR decomposition is applied to both DCT-RIWT transformed cover images and secret image blocks. The proposed method consists of two phases; section 4.1 describes the embedding phase and section 4.2 describes the extraction phase of our methods.

4.1 Embedding phase

An image of 512×512 pixels is converted to YCbCr space, and then its Y component is treated as cover image C. C is first decomposed into blocks of 16×16 pixels and the co-occurrence matrix is calculated, and according to this, the embedding map is calculated. Using the embedding map, blocks are subjected to a DCT, RIWT transformation. The secret image I of size 256×256 pixels is encrypted using a lightweight chaotic map operation in IoT sensor and sent to the home server, where it is decomposed into blocks of 8×8 pixels. QR is applied to the coefficient of every block of secret image and cover image. For every block, the R matrix of the secret image I is concealed in the R matrix of cover image C. The resulting stego image is sent to the cloud server for storage. Figure 3 shows the embedding phase and is explained below,

Fig.3

Embedding phase.

Input: Cover image C with a size of 512×512 pixels, Secret image I with a size of 256×256 pixels

Output: Stego image S with a size of 512×512 pixels, embedding map

Step 1: Cover image in RGB color space is converted to YCbCr space, and then its Y component is treated as cover image C.

Step 2: C is decomposed into blocks of size 16×16 pixels. CB_i represents the ith cover image block.

Step 3: Embedding map generated in block selection algorithm is used. The embedding map represents one if the block is used in embedding and 0 if it is not used in embedding.

Step 4: The secret image I is encrypted using a lightweight chaotic map operation at the IoT sensor node to obtain J.

Step 5: J is decomposed into blocks of 8×8 pixels. JB_i represents the ith secret image block.

Step 6: Following steps are performed for each block of CB_i whose value is 1 in the embedding map,

6.1: CB_i is transformed using DCT and 1-level RIWT to produce sub-bands: LL, LH, HL, and HH. Embedding is performed on the desired sub-band.

6.2: QR is applied to coefficient matrix CB_i of each block of cover images as follows to decompose a single matrix into two constituent matrices, $QR (sub - bandof {CB}_{i}) = {QC}_{i} x {RC}_{i}$

6.3: QR is applied to each coefficient matrix JB_i as follows to decompose a single matrix into three constituent matrices, $QR ({JB}_{i}) = {QI}_{i} x {RC}_{i}$

6.4: The R matrix of JB_i is embedded into the R matrix of CB_i of C_i as follows, $R_{i}^{1} = {RC}_{i} + α \times {RI}_{i},$ Where α is the scaling factor used to strengthen the embedding process.

6.5: The inverse of QR for step 6.2 is performed as follows to obtain a modified coefficient matrix CB_i,∥The inverse of QR (sub-band of CB_i) = QC_i x R¹ C_i

6.6: The inverse of RIWT and DCT are obtained using a modified sub-band with three other sub-bands to produce a stego image block SB_i.

Step 7: All stego image blocks SB_i are combined to produce S.

Step 8: S is the Y component of the YCbCr color space. Y is combined with other components of the YCbCr space to obtain stego image S in RGB color space.

4.2 Extraction phase

Stego image is converted to YCbCr space, and then its Y component is treated as S. Then it is decomposed into blocks of 16×16 pixels, and all block whose value is 1 in embedding map is subjected to DCT, RIWT transformation. QR decomposition is performed on the sub-bands to obtain the R matrix. The R matrix calculated in the embedding phase is used to obtain encrypted secret image block I_i. Then, all I_i blocks are merged to construct Secret image I’ with size 256×256. I’ is subjected to a lightweight chaotic map to produce secret image I. The extraction phase, as shown in Fig. 4 is explained as below,

Input: Stego image S of size 512×512, QC_i, RI_i, and embedding map.

Output: Secret image I of size 256×256

Step 1: Stego image is converted to YCbCr space, and then its Y component is treated as S.

Step 2: Stego image S is decomposed into blocks of size 16×16 pixels. SB_i represents the ith stego image block.

Step 3: Following steps are performed for each block whose value is 1 in embedding map,

3.1 SB_i is transformed using 2D DCT and 1-level RIWT to produce sub-bands: LL, LH, HL and HH and the desired sub-bands are used in the following steps.

3.2 QR is applied to SB_i as follows, $QR ({SB}_{i}) = {QI}_{i} x {RI}_{i},$

3.3 Secret image block JB_i_, is extracted as follows, $R_{i}^{2} = R^{1} I_{i} R_{i} / α$

3.4 The inverse of QR is applied to obtain coefficient JB_i. $JBi = {QC}_{i}, x R^{2} i$

Step 4: All JB_i blocks are combined to construct J.

Step 5: J is subjected to an inverse of the lightweight chaotic map to obtain secret image I.

Fig.4

Extraction phase.

5 Experimental results and analysis

In this section, the performance of our proposed method is compared with similar schemes, like those used by Chen et al. [33], Ouyang et al. [34] and Roy et al. [35]. To assess the performance of our scheme, a cover image of 512×512 pixels large, as shown in Fig. 5 (a) and three medical secret images 256×256 pixels large, as shown in Figs. 5 (b) – (d) are considered. The effectiveness of the steganographic system is determined by the imperceptibility, security, and robustness of the stego images produced by our proposed scheme. To evaluate the proposed algorithms, we also utilized a DICOM medical image dataset. These three characteristics are discussed in section 5.1 through 5.3, and the results of the comparison are discussed in section 5.4.

Fig.5

Cover and stego images.

5.1 Imperceptibility

Imperceptibility measures the quality of the produced stego images. The amount of distortion occurring in the stego images is measured using PSNR. It is computed using the following formula: $PSNR = 10 log 10 (\frac{255^{2}}{MSE})$ (1)

Where MSE denotes the mean squared error between the cover and stego images determined by the following formula: $MSE = \frac{\sum_{p = 0}^{m - 1} \sum_{q = 0}^{n - 1} (x (p, q) - (x (p, q))^{2}}{2}$ (2)

Figure 6(a) – (c) is the stego image produced by embedding the three secret images shown in Figs. 5 (b) – (d). Its PSNR values are 51.28, 50.73 and 51.45, respectively. As the PSNR values are good, we conclude that our scheme produces stego images of good quality. The histograms of these three stego images are shown in Fig. 7. All three histograms of the stego images shown in Figs. 7 (b)-(d) look alike, and moreover, look similar to the histogram of the original cover image shown in Fig. 7 (a). From these four histograms, it can be concluded that our scheme possesses very good imperceptibility.

Fig.6

Produced Stego images and its PSNR values.

Fig.7

Histogram of the cover image and its three stego image.

5.2 Security

The security of the evaluated stego images is discussed in this section. The SVM classifier is deployed for steganalysis to assess the security of our scheme. For this purpose, one thousand arbitrary images of 512×512 were considered as cover images, the secret image shown in Figs. 5(b)-(d) are embedded into it utilizing our scheme. Of the 500 pairs of cover images and their corresponding stego images, 300 pairs are used to train the classifier, and the remaining 200 pairs are used to test it. The failure rate of steganalysis on set of stego images given by our scheme is 64.37%, indicating that our scheme is secure. Implementing QR Transform and the block selection algorithm enhances the resistance to steganalysis of our scheme. Thus, we conclude that our method is secure enough to produce stego images that can counter steganalytic attacks.

5.3 Robustness

Secret images are obtained from the stego images easily when no image processing attacks have occurred. Even if a secret image is obtained from a stego image after an attack, the extraction scheme is robust. To assess the robustness of our proposed method, two measures, Normalized Cross Correlation (NCC) and Bit Error Rate (BER) are used. The stego images were subjected to the following attacks, Gaussian noise with a variance of 0.005, Salt & Pepper noise with a density of 0.01, average filter with a filter size of [3 3], median filter with a filter size of [3 3], JPEG compression with quality factor of 50%, rotation with an angle of 15 ⁰, cropping with a size of [1 1 350 350], scaling of 15%, and, finally, a combination of JPEG compression with a quality factor of 80% and Salt & Pepper noise with a density 0.01 as shown in Fig. 8. In all of the above cases, the original secret images were recovered with only minor traces of the presence of noise or the effects of image processing attacks as shown in Table 1. However, the effects of distortions by the above attacks on the stego images could not be identified using HVS. This result indicates that our scheme is robust against all possible types of attacks.

Fig.8

(a) Gaussian noise with variance 0.005 (b) Salt & pepper noise with density 0.01 (c) Average filter (d) Median filter (e) JPEG compression (50%) (f) Rotation 15% (g) Cropping (h) Scaling (i) JPEG compression 80% + Salt & pepper noise with density 0.01.

Table 1

NCC and BER values between extracted secret image and original secret image for various attacks

Attacks/NC,BER values	LL (α= 0.05)		LH (α= 0.05)		HL (α= 0.05)		HH (α= 0.05)
	NCC	BER	NCC	BER	NCC	BER	NCC	BER
Gaussian noise with M = 0, var = 0.05	0.9643	0.0553	0.9413	0.0276	0.9473	0.0377	0.9563	0.0401
Salt &pepper noise 0.01	0.9897	0.0482	0.9509	0.0385	0.9671	0.0389	0.9732	0.0384
Average filter [3 3]	0.9853	0.0573	0.9477	0.0376	0.9745	0.0331	0.9822	0.0375
Median filter [2 2]	0.9663	0.0565	0.9367	0.0316	0.9549	0.0346	0.9873	0.0356
JPEG compression (50%)	0.9857	0.0568	0.9388	0.0363	0.9635	0.0287	0.9865	0.0323
Rotation 15%	0.9462	0.0516	0.9465	0.0307	0.9651	0.0332	0.9802	0.0373
Cropping	0.9863	0.0571	0.9428	0.0332	0.9712	0.0389	0.9846	0.0376
Scaling	0.9986	0.0484	0.9413	0.03253	0.9784	0.0321	0.9825	0.0317
JPEG 80 + Salt &pepper noise 0.01	0.9575	0.0573	0.9479	0.0312	0.9639	0.0319	0.9882	0.0384

Normalized cross correlation (NCC) is computed by,

$\frac{\sum_{i = 1}^{M} \sum_{j = 1}^{N} [C_{i, j} - μ_{s}] [C_{i, j}^{'} - μ_{s}^{'}]}{\sqrt{\sum_{i = 1}^{M} \sum_{j = 1}^{N} [(C_{i, j} - μ_{s})]^{2} \sqrt{\sum_{i = 1}^{M} \sum_{j = 1}^{N} (C_{i, j}^{'} - μ_{s}^{'})^{2}}}}$ (3)

Bit Error Rate (BER) is computed by, $\frac{Number of error bits in a image}{Total number of bits in a image}$ (4)

5.4 Comparison

Our scheme is compared with the other three related schemes as shown in Table 2. The NCC results between the original secret images and the extracted secret images from our scheme for almost all the nine parameters are superior over the other three schemes. The experimental results in Table 2 and the graphical analysis in Fig. 9 show the superiority of our scheme.

Fig.9

Comparison of our scheme with other three related schemes.

Table 2

Comparison of NCC values of our scheme with other three related schemes

Attacks/Schemes	[33]	[34]	[35]	Ours
Gaussian noise with 0.005	0.6005	0.8460	0.9453	0.9643
Salt &pepper noise 0.01	0.7488	0.9388	0.9888	0.9897
Average filter	0.9245	0.9618	0.9757	0.9853
Median filter	0.9408	0.8742	0.9532	0.9663
JPEG compression (50%)	0.9999	0.8040	0.9812	0.9857
Rotation 15%	0.7927	0.8542	0.9351	0.9462
Cropping	0.9851	0.9779	0.9464	0.9863
Scaling	0.9970	0.9358	0.9412	0.9986
JPEG 80 + Salt &	0.7454	0.7864	0.9367	0.9575
pepper noise 0.01

6 Conclusion

Due to unexpected developments in mobile and communication devices, IoT is now widely used to resolve serious challenges in the context of healthcare and critical remote monitoring systems. The applicability of IoT can only be achieved to the fullest only when its inherent risks of disclosure to unauthorized users and the associated effects are minimized and dealt with seriously. This paper suggests using a twin layer security measure, one at the IoT sensor device end and the other at the home server, to minimize security risks. In IoT sensor device, lightweight chaotic map operation is carried out due to its low computational power and the combined approach of cryptography and steganography is performed at the home server as it has the more computational capability. Because of the shift invariant and redundancy property of RIWT, the embedding capacity is increased, and the extraction process is also precise. Achieving successful steganography depends on the development of improved algorithms by choosing an appropriate location. In our scheme, the cover image is divided into blocks, and the best block is chosen using a co-occurrence matrix. Finding an optimal block for embedding not only minimizes distortion of image quality but also improves embedding efficiency and can increase the resistance to steganalysis. The results of testing on three parameters indicate that our proposed scheme performs decently for safeguarding confidential image transmission through IoT. In the future, we propose a scheme where all the three planes of the RGB color space can be used for embedding, instead of converting cover images in the RGB color space into the YCbCr color space and considering only its Y component for embedding.

References

Darwish

, Hassanien

A.E.

, Elhoseny

, Sangaiah

A.K.

and Muhammad

, The impact of the hybrid platform of internet of things and cloud computing on healthcare systems: Opportunities, challenges, and open problems, Journal of Ambient Intelligence and Humanized Computing (2017), 1–16. https://org/10.1007/s12652-017-0659-1.

Nitti

, Girau

and Atzori

, Trustworthiness management in the social internet of things, IEEE Transactions on knowledge and Data Engineering26(5) (2014), 1253–1266.

Jang

, Jung

I.Y.

and Park

J.H.

, An effective handling of secure data stream in IoT, Applied Soft Computing68 (2018), 811–820.

Kumari

, Tanwar

, Tyagi

, Kumar

, Maasberg

and Choo

K.K.R.

, Multimedia big data computing and Internet of Things applications: A taxonomy and process model, Journal of Network and Computer Applications (2018), 169–195.

Chasaki

and Mansour

, Security challenges in the internet of things, International Journal of Space-Based and Situated Computing5(3) (2015), 141–149.

Katzenbeisser

and Petitcolas

, Information hiding techniques for steganography and digital watermarking, Artech house (2000), 1–3.

Tiwari

and Shandilya

D.M.

, Evaluation of various LSB based methods of image steganography on GIF file format, International Journal of Computer Applications (2010), 1–4.

Provos

and Honeyman

, Hide and seek: An introduction to steganography, IEEE Security & Privacy99(3) (2003), 32–44.

Das

and Das

, Secure data transfer in IoT environment: Adopting both cryptography and steganography techniques, Research in Computational Intelligence and Communication Networks (ICRCICN), 2016 Second International Conference on. IEEE, (2016), 296–301.

10.

Kharrazi

, Sencar

H.T.

and Memon

, Cover selection for steganographic embedding, 2006 International Conference on Image Processing. IEEE, (200) 1–4.

11.

Yehia

, Khedr

and Darwish

, Hybrid security techniques for Internet of Things healthcare applications, Advances in Internet of Things5(03) (2015), 21–25.

12.

Sreekutty

M.S.

and Baiju

P.S.

, Security enhancement in image steganography for medical integrity verification system. Circuit, Power and Computing Technologies (ICCPCT), 2017 International Conference on. IEEE, (2017), 1–5.

13.

Bashir

, Hasan

A.S.B.

and Almangush

, A new image encryption approach using the integration of a shifting technique and the AES algorithm, International Journal of Computers and Applications42(9) (2012), 38–45.

14.

Abdelaziz

, Elhoseny

, Salama

A.S.

and Riad

A.M.

, A machine learning model for improving healthcare services on cloud computing environment, Measurement119 (2018), 117–128.

15.

Yin

J.H.J.

, Fen

G.M.

and Mughal

, Internet of Things: Securing Data using Image Steganography. Artificial Intelligence, Modelling and Simulation (AIMS), 2015 3rd International Conference on. IEEE (2015), 310–314.

16.

Anwar

A.S.

, Ghany

K.K.A.

and Mahdy

H.E.

, Improving the security of images transmission, International Journal3(4) (2015), 7–13.

17.

Khalil

M.I.

, Medical image steganography: Study of medical image quality degradation when embedding data in the frequency domain, International Journal of Computer Network and Information Security9(2) (2017), 22–28.

18.

Abawajy

J.H.

and Hassan

M.M.

, Federated internet of things and cloud computing pervasive patient health monitoring system, IEEE Communications Magazine55(1) (2017), 48–53.

19.

Osborne

Chapter 42-internet of things and cloud computing, Geng, H. (ed.) Part VI: Cloud, Legal, Innovation, and Business Models in Internet of Things and Data Analytics Handbook, 1st edn. (2017), 683–698.

20.

Rahman

S.M.M.

, Hossain

M.A.

and Hassan

M.M.

, Secure privacy vault design for distributed multimedia surveillance system, Future Generation Computer Systems55 (2016), 344–352.

21.

Gilbert

, Privacy and security legal issues, Internet of Things and Data Analytics Handbook (2017), 699–718.

22.

Al Alkeem

, Shehada

, Yeun

C.Y.

, Zemerly

M.J.

and Hu

, New secure healthcare system using cloud of things, Cluster Computing20(3) (2017), 2211–2229.

23.

Sadek

R.A.

, SVD based image processing applications: State of the art, contributions and research challenges. arXiv preprint arXiv:1211.7102 (2012), 26–34.

24.

Makbol

N.M.

and Khoo

B.E.

, A new robust and secure digital image watermarking scheme based on the integer wavelet transform and singular value decomposition, Digital Signal Processing33 (2014), 134–147.

25.

Guo

J.M.

and Prasetyo

, Security analyses of the watermarking scheme based on redundant discrete wavelet transform and singular value decomposition, AEU-International Journal of Electronics and Communications68(9) (2014), 816–834.

26.

Subhedar

M.S.

and Mankar

V.H.

, Image steganography using redundant discrete wavelet transform and QR factorization, Computers & Electrical Engineering54 (2016), 406–422.

27.

Gunjan

, Mitra

and Gaur

M.S.

, Contourlet based image watermarking scheme using Schur factorization and SVD, International Conference on Advances in Communication, Network, and Computing. Springer, Berlin, Heidelberg (2012), 337–340.

28.

Janakiraman

, Thenmozhi

and Rayappan

J.B.B.

, Lightweight chaotic image encryption algorithm for real-time embedded system: Implementation and analysis on 32-bit microcontroller, Microprocessors and Microsystems56 (2018), 1–12.

29.

Sajedi

and Jamzad

, Using contourlet transform and cover selection for secure steganography, International Journal of Information Security9(5) (2010), 337–352.

30.

Haralick

R.M.

and Shanmugam

, Textural features for image classification, IEEE Transactions on Systems, Man, and Cybernetics6 (1973), 610–621.

31.

Zhang

, Wu

, Gao

, Sun

and Yan

, Robust reversible watermarking algorithm based on RIWT and compressed sensing, Arabian Journal for Science and Engineering43(2) (2018), 979–992.

32.

Rabie

and Kamel

, High-capacity steganography: A global-adaptive-region discrete cosine transform approach, Multimedia Tools and Applications76(5) (2017), 6473–6493.

33.

Chen

, Coatrieux

, Chen

and Sun

, Full 4-D quaternion discrete Fourier transform based watermarking for color images, Digital Signal Processing28 (2014), 106–119.

34.

Ouyang

, Coatrieux

, Chen

and Shu

, Color image watermarking based on quaternion Fourier transform and improved uniform log-polar mapping, Computers & Electrical Engineering46 (2015), 419–4.

35.

Roy

and Pal

A.K.

, An SVD based location specific robust color image watermarking scheme using RDWT and Arnold Scrambling, Wireless Personal Communications98(2) (2018), 2223–2250.