ViMDH: Visible-Imperceptible Medical Data Hiding for Internet of Medical Things

Over the recent years, volume of medical images and related digital records, called electronic medical records, generated, shared, and stored by different intelligent devices, sensors, and Internet of medical things networks, to name a few, has drastically increased. Such records are shared by cloud providers for storage and further processing. However, an increasingly serious concern is the illegal copying, modification, and forgery of medical records. This article presents a visible and imperceptible medical data hiding technique, namely ViMDH, which can prevent to intellectual property theft of medical records. The carrier image is visibly marked with logo mark, which is suitable for owner identification and avoid illegal duplication, and then an imperceptible data hiding based on nonsubsampled shearlet transform (NSST), redundant discrete wavelet transform (RDWT), and multiresolution singular value decomposition is introduced. Finally, key-based encryption scheme designed by RDWT-RSVD ensure the security of the watermarking system. Under the experimental evaluation, our ViMDH is not only visible and imperceptible, but also has a satisfactory advantage in robustness and security compared with the traditional watermarking schemes.


I. INTRODUCTION
R ECENTLY, the Internet of medical things (IoMT) based physical objects are interconnected with sensors to collect, process, and store EMRs for smarter decisions in the cloud computing environment. Medical images are the information carriers that play a major role in smart healthcare communication [1]. Furthermore, in the past few years, volume of EMRs have been continuously generated, transmitted, and stored online among various medical specialists due to the appearance of COVID-19 pandemic situation. According to [2], a significant growth of 154% in the telehealth visits is experienced in March 2020 when compared with March 2019. Also, cloud-based healthcare applications have ensured uninterrupted clinical diagnosis with mobility support and low latency [3]. It is no doubt that advancement in healthcare industries has brought great continence for communication, but has also caused serious infringement of EMRs. Watermarking is considered as a powerful tool to prevent copyright violation of medical records, which involves embedding valuable information into carrier media as a form of identification [4]. In the past few years, researchers have adopted watermarking scheme supported by encryption to provide stronger security of EMRs data for smart healthcare applications [3], [5]. Based on human visual perception, there are usually two ways for mark embedding in watermarking systems, namely visible and imperceptible [4]. Visible watermarks are transparent, which are imprinted on the cover media, including both images and videos. These marks can be recognized through visual means, hence, are poorly imperceptible in nature. They provide ownership verification and avoid the illegal duplication and forgery. On the contrary, in invisible watermarking, a secret data are concealed in the cover image through certain logical processing, which can only be determined by genuine user. These are robust in nature and provide confidentiality and maintain the privacy of the exchanged data. It also prevents detachment of patient's data from their respective medical images. However, it may deteriorate the visual quality of the cover media. Therefore, researchers have introduced watermarking techniques to effectively solve the copyright violation of medical records.
For example, a hybrid of robust and visible watermarking for DICOM images is presented by Mata et al. [6]. The authors followed visible marking in spatial domain of the logo image to provide authentication. Furthermore, hash based robust marking in DCT domain is used to prevent the detachment of patient data and their respective medical image. Furthermore, visible watermarking method in the region of noninterest (RONI) is developed by Thanki et al. [7] to authenticate the digital medical records. This technique uses human visual system to find the RONI of the medical image. Different image processing operations including addition, subtraction, complement, and cropping are used to perform the visible marking of the mark. Kahlessenane et al. [8] developed a robust and blind watermarking method for the medical images where the watermark bits are concealed in upper triangular Schur matrix. The authors also compared the performance of different transforms including discrete wavelet transform (DWT), discrete cosine transform (DCT), NSST, and nonsubsampled contourlet transform when 1551-3203 © 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
combined with Schur decomposition. DWT-Schur and NSST-Schur based methods shows best visual quality and robustness, respectively. Another robust MSVD based watermarking method is proposed by Akhbari and Ghaemmaghami [9]. The use of entropy masking for embedding and extraction procedure makes the proposed method adaptive with high visual quality and resistance to attacks. The comparative analysis shows better performance in comparison to traditional DWT based watermarking. A dual watermarking algorithm supported with compression-beforeencryption is introduced in [10]. The EPR data is encoded by turbo-code before concealing into the cover to reduce the error rate. Furthermore, dual marks are conceals using RDWT and RSVD transforms in RONI of the carrier image. The final marked image is compressed then encrypted to ensure better compression performance and security over open network. Singh et al. [11] studied the effect of different error correcting codes on the robustness and invisibility of the DWT-SVD based watermarking scheme. Encoded text and segmented image mark are imperceptibly inserted within the transformed cover image. Result analysis concluded that hybrid of BCH and repetition codes have higher resistance against noises and attacks. Priya et al. [12] developed an encryption based watermarking technique to ensure the security of medical images. It uses hybrid of Paillier Cryptosystem, DWT, and DCT to invisibly embed the mark within the encrypted host image. The security and authentication process are further improved by applying poker shuffling transformation on the mark image. Anand and Singh [13] investigated the effectiveness of watermarking scheme for COVID-19 images and explored its impact on different attacks. In their article, multiple-marks are embedded by using the different transformed-domain scheme for various kinds of images. Although the robustness of the scheme has been greatly improved in this article, the security of the image is still weak. Curvelet transform based watermarking method for recovery of ROI is proposed by Eswaraiah et al. [14]. Initially, ROI of medical image is selected and its hash value is calculated using SHA-1 for better authentication. This hash value along with the binary form of ROI region is concealed in the curvelet coefficient of RONI region, which can later be extracted to recover the ROI region.
Contributions: Although the reviewed methods provide authenticity and robustness to common attacks, they have limited embedding capacity with low security. Also, these watermarking methods are essentially designed to individually visible or imperceptible domain, ignoring the advantages of a single type of domain. In this article, a visible and imperceptible medical data hiding technique, namely ViMDH, that can prevent the infringement of medical records. Major novelties of this article are as follows.

1) Advantages of visible and imperceptible watermarking:
Combination of visible and imperceptible watermarking effectively provides ownership identification and robustness of the watermark by avoiding the illegal duplication and forgery of the EMRs data. It also prevents detachment of patient's data from their respective medical images. 2) High quality of the marked and extracted mark: Combination of NSST-RDWT-MSVD transformed scheme is used to perform imperceptible marking of different marks within the visible marked carrier media. This combination offers not only the high robustness and imperceptibility, but also increases the mark capacity with linear complexity [9], [15]. 3) Strong security: Hash value of visibly marked carrier image is calculated using RDWT-RSVD based hash generation method to provide better authentication. The generated hash value is eventually placed in the DWT subband of the image mark. Also, the hash value is encoded with Turbo code to reduce the amount of error, if any. Additionally, key-based encryption scheme designed by RDWT-RSVD ensure the security of the watermarking system. 4) Better robustness performance: Compared with the traditional schemes, the proposed ViMDH technique has better robustness, indicating its potential for secure IoMT-based healthcare. The rest of this article is organized as follows. Detailed explanation of visible marking, hash value generation, watermark generation, embedding and extraction of dual marks, and encryption of marked image is discussed in Section III. The experimental analysis is done in Section IV. Finally, Section V concludes this article.

II. VIMDH TECHNIQUE
The simplified visible and imperceptible ViMDH mechanism is shown in Fig. 1. The entire ViMDH mechanism is expressed as follows: 1) visible marking, 2) hash value computation of marked carrier media, 3) mark generation for embedding, 4) embedding and recovery of generated marks, and 5) encryption of marked data. The stepwise procedure of each phase is illustrated in Algorithm 1 to Algorithm 5, respectively. Some commonly used notations in algorithms are listed in Table I.

A. Visible Marking
The simplified procedure of visible marking for the purpose of ownership verification is shown in Fig. 2. In this procedure, the input carrier image "Ḉ" is visibly concealed with the hospital logo "v_mark". Initially, "Ḉ" and "v_mark" are preprocessed by adjusting the size and conversion to gray scale. Furthermore, the co-ordinates to place the visible watermark are adjusted and temporary image "temp", is formed using the finalized co-ordinated. Finally, "temp" is visibly concealed in the pre-processed carrier image using opacity factor "vis_ α", by applying the following equation: The detailed procedure of visibly marking the carrier image is described in Algorithm 1.

B. Hash Value Computation of Visibly Marked Media
The simplified procedure of hash value computation of visibly marked media for the purpose of better authentication is shown in Fig. 3. The normalization procedure is followed to ensure invariant property against geometric modifications [16]. The standard image "standard_vḈ", holds the relevant information of marked carrier image "vis_Ḉ", and is computed using geometric and central moments, which serves as the input arguments for the normalization procedure. The geometric moment "g xy " of       Here, "s" and "a x b y" denotes the support of "GI (a, b)" and corresponding basic set, respectively. Furthermore, the central moments "c xy " can be computed as where x, y = 0, 1, 2, . . . .
Here, "t α " and "t β " are calculated as

C. Mark Generation for Embedding
In the next phase, the dual watermarking is realized by concealing hash value of visibly marked image, "hash", within the gray scale medical image, "w_img" (see Fig. 4). The text mark, "hash", is first encoded with Turbo Code to minimize the effect of channel noise. Furthermore, DWT is applied on the carrier image producing the following sub-bands "wA," "wV," "wH," and "wD". Encoded mark, "en_hash" is placed inside "wD" sub-band using gain factor "text_α". Finally, inverse DWT is applied to generate the final watermark "fw_img". The detail of dual watermark generation is discussed in Algorithm 3.

E. Encryption of Marked Media
As shown in Fig. 6, key-based encryption is applied on the final marked data "W_Ḉ" to obtain the encrypted marked image "enc_WḈ" before circulated online. First, "W_Ḉ" is scrambled using random permutation to form scrambled image "sh_WḈ". Furthermore, combination of RDWT and RSVD transforms are carried out for "W_Ḉ", resulting in singular matrices, including "kU1," "kS1," and "kV1". It is followed by fusion of "kU1" and "kV1" to form a new matrix "B", which is again decomposed using RSVD resulting in "kU2," "kS2," and "kV2". The product of "kU2" and "kV2" gives the final key "key". Finally, scrambled image, "sh_WḈ" and "key" are used to generate the Algorithm 4: Embedding and Recovery of Generated Mark.

III. RESULTS AND ANALYSIS
In this section, the performance of the proposed ViMDH technique is evaluated by applying various attacks to the marked image. All the experiments were performed using MATLAB 2021b. The tested mark data includes hospital logo, medical  image, and hash value of size 64×64, 256×256, and 110 bits, respectively. Three hundred 300 covid-19 images [17] and fifty other images [18] with a size of 512×512 are used as carrier images. The process of visible marking consumes 0.0016 s to generate the visibly marked image. Furthermore, NSST-RDWT-MSVD based watermarking method utilizes 0.7642 and 0.2656 s for embedding and extraction of dual marks, respectively. Finally, time taken by encryption and decryption procedures are 0.0238 and 0.5718 s, respectively. A number of measure indices are required in order to test the performance of the ViMDH technique. The similarity evaluation of among two images is computed by PSNR and SSIM [19]. Also, hamming distance is used for evaluating the robustness of the hash values as obtained by Algorithm 2. NC and BER score are commonly used for evaluation the robustness performance against attacks. The NPCR and UACI are two well-known indicator for strength of the encryption technique used [19]. Finally, the subjective evaluation includes the histogram comparison of the visibly and imperceptibly marked carrier images.
The effect of gain factor (α), ranging between 0.01 to 1, is studied in Table II. With the minimum gain value of 0.01, best PSNR and SSIM values are 96.2956 dB and 0.9999, respectively. Highest NC = 1 is observed at gain value = 1. With BER = 0 for all cases, the text is recovered exactly with no error. The results summaries good NC score at high gain factor and high PSNR at relatively lower values of gain factor. Also, values of NPCR and UACI shows strong security ability of encryption scheme. Value of gain factor is considered as "0.5" for further evaluations. The difference between the histogram distribution of visibly marked and imperceptibly marked carrier images is shown in Fig. 7. The result shows high similarity between in the gray histogram distribution of the two considered images. We also recorded the quantitative score of objective analysis from various medical images and other grayscale images in Table III. With α = 0.5, the average values of PSNR and SSIM for medical images are 75.8615 dB and 0.9833, respectively. Also, mean value of NC = 0.9927 and BER = 0 demonstrates high robustness. The acceptable scores of NPCR and UACI are obtained for the proposed encryption technique. Furthermore, the performance, in terms of hamming distance score, of the proposed hash value generation is compared with the traditional DWT-SVD based method [16] in Fig. 8. Notably, an improvement in the hamming score is experienced due to the better robustness offered by the hybrid of RDWT and RSVD.
The performance of our method is investigated after imposing different kinds of attacks in Table IV. The maximum and minimum NC results are recorded as 0.9939 and 0.9726, respectively. Also, the value of BER is obtained as "0" expect for rotation and high intensity translation attack. Finally, comparative assessment of the proposed work with similar schemes [3], [8], [20] has been demonstrated in Fig. 9. Also, Table V shows the comparison of embedding capacity and visual quality of the proposed work with other published work [3], [8], [20]. The fusion of NSST-RDWT-MSVD for hiding dual watermarks in the cover image helped in scoring maximum improvement of 47.44% over the traditional work. Overall, the implementation of visible-imperceptible watermarking for hiding multiple marks  and efficient encryption improved the identity verification, invisibility, robustness, and security criteria in our method.

IV. CONCLUSION
This article presented a visible and imperceptible medical data hiding technique, together with encryption technique for the prevention of the infringement of medical records. First, the carrier image was visibly marked with logo mark, which was suitable for ownership verification. Second, hash value of the visibly marked carrier image was computed using RDWT-RSVD for better authentication. Third, NSST, RDWT, and MSVD were then employed to conceal the generated mark into the visible media. Finally, key-based encryption scheme designed by RDWT-RSVD ensure the security of the watermarking system. The experimental results indicated that the proposed ViMDH technique was characterized by good imperceptibility, strong security, remarkable payload, and high robustness against attacks. Compared with traditional watermarking methods, the ViMDH has better performance to protect the copyright of EMRs and related medical images that will be transmitted over the open networks. However, the computational cost of this article can further be reduced. Our future direction is to design time and cost-efficient watermarking scheme in machine/deep learning environment.