Robust transmission of medical records using dual watermarking and optimization algorithm

Nowadays, security of medical records is an important issue for tele-health system. Motivated by importance of such critical issue, we propose a robust watermarking method in the DWT-SVD-optimization domain. To solve the security issue, traditional watermarking schemes uses manual scaling factor to manage balance between imperceptibility, robustness and capacity. However, selection of manual scaling factor loses their optimize trade-off between these important parameters of watermarking. The suggested scheme encourages protection of medical data via techniques of dual watermarking and optimization scheme. In our scheme, the data owner imperceptibly embeds the dual watermarks in the medical cover image for extra level of security. Here, appropriate optimization schemes are used to find the scaling factor for embedding purpose. Further, the performance of this scheme is examined and compared. Moreover, the patient text data is coded via hamming code before inserting in to the cover so that bit error rate can be avoided or eliminated, if any. We show that the suggested scheme not only offers the high imperceptible but also robust for various attacks. Compared with existing schemes, our work offers more robustness while imperceptible and good capacity at the same time.


Introduction
In tele-health, there has been an exponential increase in the communication of medical images through open network systems [1]. These services use information and communication technology (ICT) tools for effective storage, management and communication of the medical records for various purposes [2]. While transferring the medical images might bring many issues including being tampered by intruders, identity theft problem and privacy leakage. Further, all the medical records are transmitted using Digital Imaging and Communication in Medicine (DICOM) standard [3]. It can be seen that this standard is not much effective for reliable healthcare system [2] [3]. Additionally, identity theft and privacy leakage of healthcare data is a growing crime which results in a huge damage to finances of the individual [4]. Presently, a lot of significant patient information is stored in the local server of the medical centre and distributed over the network in Coronavirus periods [5]. On the other hand, this may lead to high risk of data security and privacy in the current advanced healthcare systems [6]. To overcome such legal and ethical issues, watermarking is used to secure medical contents [1]. It facilitates embedding sensitive information into digital content such that it can later be extracted by authorized party only [4]. Notably, robustness, invisibility and capacity are the primary requirements of any watermarking method [1], which is difficult to balance the trade-off between these requirements [7]. Wavelet based adaptive and robust watermarking approach is proposed by Gangadhara et al. [22] for securing the healthcare applications. The cover is decomposed using modified DWT and SVD. Further, PSO is implemented to optimize the scaling factor, which is then used to embed the watermark into region with maximum entropy.

Design of the proposed scheme
Our scheme mainly comprises of the following phases, a. Embedding and recovery of the concealed watermarks b. Optimization of the scaling factor. Forthcoming subsections explain both the phases in details. Figure 1 illustrates the concealing of the multi-watermarks and optimizing the scaling factor (SF) using ABC algorithm.

Embedding and extraction of multi-watermarks
Listing 1 describes the embedding process in detail. In this process, second level DWT-SVD is applied on the cover image. Prior to embedding, the image watermark is sub divided into two equal parts, making the technique more imperceptible [23]. After, and sub-images of the image watermark are concealed using optimized SF (refer listing 2) in the resultant singular component of the specified subbands of the cover. Additionally, hamming code is utilized to encode the patient watermark (character form) so that bit error rate can be avoided or eliminated, if any.
The encoded bits of the patient watermark are hidden in the higher-level sub-band of the cover. Finally, inverse of SVD-DWT is utilized to obtain the final marked image. For extracting the watermarks, exact reverse embedding procedure is to be followed.
Step 3. Determine singular value of LH1 and HL1 components.
Step 4. Read watermark image (w_img) of size 256×256 and divide it equal parts Step 5. Embed w_img1 and w_img2 into S LH and S HL sub-bands using gain factor α.
Step 6. Apply inverse SVD to attain the marked LH1 and HL1.
Step 7. Read w_text and convert it into ASCII format.
Where, len is the length of w_final. Step 11. final_img is obtained via inverse DWT.

Optimization of the scaling factor
Any general watermarking system need to maintain the balance between two major characteristics, namely imperceptibility and robustness [7]. It can be seen that the larger and smaller value of SF offer good robustness and imperceptivity, respectively [2]. So, an optimal value of SF is required to maintain a trade-off between the two. In this context, ABC algorithm [24] is used to find the optimal value of the SF for embedding and extraction process of the mark data. Fig. 2 shows the detail working process of ABC optimization scheme. It has been proven that the suitable range for a confined initialization of random swarms is 0.005-0.06 [21]. The basic optimization parameters for the scheme are illustrated in Table 1. Further, fitness function is defined as [25]: Step 1: Determine PSNR, NC and BER values without attack as illustrated Listing 1.
Step 2: Determine NC and BER values with n-attacks.

Results
Experimental analysis of the proposed work is done using MATLAB R2017a on a 64-bit, 2.10 GHz processor, 8 GB RAM system. The experimental parameters are: size of the cover medical image is 512×512 [26], and patient report and medical image are considered as mark data of size 11 characters and 256×256, respectively. Robustness of the proposed watermarking approach is verified against different attacks listed in Table 2. PSNR and NC are used to estimate the distortion and robustness, respectively. On the other hand, BER is used to calculate the error (if any) in received bits i.e. for text watermark. Further, optimized SF value is determined via ABC algorithm, which makes the good balance between tradeoff parameters.
The PSNR, NC and BER value against considered attacks are recorded in Table 3, where the suggested technique obtained satisfactory performance. It can be seen from the table, NC value is greater than 0.7, BER is zero (except for atck_5 (cropping attack)), and PSNR value is 35.8651 dB. Further, NC and BER value for varying cover images and attacks are recorded in Table 4. It can be recorded that maximum values of PSNR and NC are 39.9559 dB and 0.9999, respectively for cell image. Refer Table 4, the BER is zero (except for atck_5 (cropping attack)) under the consideration. The results show that the proposed technique is robust and shows versatility since it offered virtuous results when tested with a variety of cover images. Furthermore, the effect of number of iterations under ABC optimization on PSNR, NC and BER value are recorded in Table 5. We noticed that value of SF approaches the optimal value as the number of iterations increases. It has been shown that performance of the system depends of number of iterations. The effect of optimization algorithms (PSO, firefly, and ABC) on NC performance is shown in Figure 3. Notably, ABC provides better results against under consideration. Overall, the objective assessment of the proposed scheme offered good results and also outperformed when compared with similar approach [2]. As shown in this Figure  4, the performance of suggested method is always better except atck_5 (cropping attack).

Conclusion
Based on optimization technique, we developed a non-blind robust watermarking in the DWT-SVD domain. In our scheme, dual watermarks are imperceptibly embedded within the singular matrices of the selected DWT sub-bands of the cover for extra level of security and authenticity. The method uses optimization techniques to make the balance between robustness, imperceptivity and capacity. Out of the three different optimizations techniques, ABC algorithm performed better than PSO and firefly scheme. Moreover, the patient text data is encoded via hamming code before embedding in to the cover so that bit error rate can avoid or eliminate, if any. Our results analysis has indicated that the proposed scheme is efficient for tele-health applications. In future, we will study the performance of proposed scheme with some more appropriate transform and hybrid meta-heuristic technique, which will offer better results. We may also incorporate the concepts of machine learning, neural networks and blockchain to further enhance the performance.

Comparison of BER Value
Proposed method Technique in [2]