Securing medical data while transmission on the network is required because it is sensitive and life-dependent data. Many methods are used for protection, such as Steganography, Digital Signature, Cryptography, and Watermarking. This paper introduces a novel robust algorithm that combines discrete wavelet transform (DWT), discrete cosine transform (DCT), and singular value decomposition (SVD) digital image-watermarking algorithms. The host image is decomposed using a two-dimensional DWT (2D-DWT) to approximate low-frequency sub-bands in the embedding process. Then the sub-band low-high (LH) is decomposed using 2D-DWT to four new sub-bands. The resulting sub-band low-high (LH_{1}) is decomposed using 2D-DWT to four new sub-bands. Two frequency bands, high-high (HH_{2}) and high-low (HL_{2}), are transformed by DCT, and then the SVD is applied to the DCT coefficients. The strongest modified singular values (SVs) vary very little for most attacks, which is an important property of SVD watermarking. The two watermark images are encrypted using two layers of encryption, circular and chaotic encryption techniques, to increase security. The first encrypted watermark is embedded in the S component of the DCT components of the HL2 coefficients. The second encrypted watermark is embedded in the S component of the DCT components of the HH_{2} coefficients. The suggested technique has been tested against various attacks and proven to provide excellent stability and imperceptibility results.

Securing medical images is an important issue in healthcare. Patient information such as electronic health records (EHR) and medical images, e.g., X-rays, computerized tomography (CT), and Magnetic resonance imaging (MRI) scans, need to be shared on the network [

Transform domains are used in watermarking techniques [

A discrete cosine transform (DCT) converts images into low, mid, and high-frequency sub-bands. While high-frequency areas of the image can be removed when attacks (compression and noise) are applied, the image’s major and important viewable regions are found in the low-frequency sub-bands [

In discrete wavelet transform (DWT), an image is split into four non-overlapping multi-resolution sub-bands [

Using the singular value decomposition (SVD) transformation, a matrix can be split into three identically sized matrices. An image is a matrix of nonnegative scalar elements from the perspective of linear algebra [

Encryption is a series of mathematical operations applied to data to generate a different type of data called a cipher [

Image pixels are randomly reconfigured in chaotic image encryption. The Cat map, the Line map, and the Baker map are all examples of chaotic maps that can be utilized for picture encryption [

In this work, we develop a watermarking scheme to assess the effects of several medical image-processing techniques as attacks like Gaussian noise. A DWT-DCT-SVD transform method was performed to embed the watermark. The secret image is scrambled using circular and chaotic encryption. Simulation using Matlab was used to apply various image processing techniques to insert and extract watermarks to check whether watermarking is effective.

This section summarizes the recently developed watermarking techniques introduced for medical imaging. In [

In [

In [

A watermarking method with two modules, embedding, and extraction, was proposed by Aparna et al. in [

The proposed technique applies DWT of the original image to LL, LH, HL, and HH. Then the sub-band LH is decomposed using 2D-DWT to LL_{1}, LH_{1}, HL_{1}, and HH_{1}. Then the sub-band LH1 is decomposed using 2D-DWT to LL_{2}, LH_{2}, HL_{2}, and HH_{2}. Two frequency bands (HH_{2}) and (HL_{2}) are transformed by using 2D-DCT and then applying SVD to the DCT coefficients to get S values, and then the S values are added to the watermark at different depths. The two watermark images are encrypted using circular encryption and chaotic to increase security. The first encrypted watermark is embedded in the S component of the HL2 Coefficients’ DCT components. In contrast, the second encrypted watermark is embedded in the S component of the HH2 Coefficients’ DCT components.

Then the method is applied by using another technique by embedding the watermark in the least significant bit of the singular values. Inverse SVD on changed S vector and original U, V vectors are used to create the watermarked image, which is then processed twice using inverse 2D-DCT and inverse 2D-DWT. This proposed addition technique improves the robustness of the host image without degrading it. Watermark embedding at the sender and watermark extraction at the receiver are the two aspects of the proposed technique, as introduced in algorithm 1.

Two independent parameters can be used to assess the quality of digital watermarking: imperceptibility and resilience. The PSNR of the host picture and the embedded image in dB are used to determine imperceptibility. Higher PSNR is preferred since it effectively hides the designated picture. The original and restored watermark images are compared to determine robustness. When the peak signal-to-noise ratio (PSNR) is high, the watermarked image resembles the original image more closely, implying that the watermark is undetectable. Watermarked images with a PSNR greater than 35 are generally acceptable. The following performance evaluation measures are used in most watermarking projects.

For invisible watermarking methods, the watermark should be imperceptible, and the human eye should not be able to distinguish between the watermarked and the original images. This measure is subjective and thus is not always reliable for evaluating a watermarking algorithm.

The peak signal-to-noise ratio (PSNR) measure is used between the watermarked and the unwatermarked images. It is related to imperceptibility, where a higher PSNR means a higher imperceptibility [

A correlation coefficient measure is used between the extracted and the original watermarks, where a higher correlation coefficient means that the extracted watermark is the one of interest. This measure is calculated as follows [

where, W and W^{^} are the original and extracted watermarks, respectively.

Experiments have been done on both medical and standard images. The test shows the imperceptibility of the original image, which has been watermarked. The robustness of the watermark is approved by applying different types of attacks: image cropping, noise, i.e., salt and pepper-Gaussian, image rotation, and median filter. Experiments were done on different images; a 1024 × 1024-host image and 128 × 128 of two watermark images were used.

The extracted watermarks with and without attacks such as median filter, Rotation, Cropping, Gaussian, and Salt and Pepper are shown in

A PSNR representation with values higher than 35 dB is within an acceptable level of degradation, which means that it is almost not seen by the Human visual system (HVS). The extracted watermarks after various attacks with Correlation values as a measure for robustness are shown in

(A) Results of watermarking using DWT-DCT-SVD-additive technique with watermark depth = 0.09 | (B) Results of watermarking using DWT-DCT-SVD-LSB | ||||||
---|---|---|---|---|---|---|---|

Attack type | PNSR dB | Correlation of watermarked 1 | Correlation of watermarked 2 | PNSR dB | Correlation of watermarked 1 | Correlation of watermarked 2 | Attack type |

Without attack | 38.86 | 0.99 | 0.99 | 64.97 | 0.99 | 0.84 | Without attack |

Median filter [3 3] | 42.32 | 0.97 | 0.95 | 59.66 | 0.34 | 0.83 | Median filter [3 3] |

Gaussian noise variance = 0.1 | 12 | 0.99 | 0.84 | 12.01 | 0.42 | 0.47 | Gaussian noise variance = 0.1 |

Rotation 80 | 15 | 0.85 | 0.86 | 14.42 | 0.30 | 0.35 | Rotation 80 |

Cropping | 17.43 | 0.98 | 0.98 | 17.45 | 0.88 | 0.89 | cropping |

Salt and pepper noise variance = 0.02 | 19.54 | 0.99 | 0.88 | 24.43 | 0.42 | 0.47 | salt and pepper noise variance = 0.02 |

(A) Results of watermarking using the DWT-DCT-SVD-additive technique with watermark depth = 0.09 | (B) Results of watermarking using DWT-DCT-SVD-LSB | ||||||
---|---|---|---|---|---|---|---|

Attack type | PNSR dB | Correlation of watermarked 1 | Correlation of watermarked 2 | PNSR dB | Correlation of watermarked 1 | Correlation of watermarked 2 | Attack type |

Without attack | 37.34 | 0.97 | 0.97 | 69.6 | 0.95 | 0.91 | Without attack |

Median filter [3 3] | 39.83 | 0.73 | 0.87 | 45.41 | 0.92 | 0.56 | Median filter [3 3] |

Gaussian noise variance = 0.1 | 11.36 | 0.95 | 0.91 | 11.38 | 0.42 | 0.25 | Gaussian noise variance = 0.1 |

Rotation 80 | 11.27 | 0.59 | 0.41 | 10.68 | 0.59 | 0.63 | Rotation 80 |

Cropping | 10.32 | 0. 95 | 0.95 | 10.33 | 0.88 | 0.89 | cropping |

Salt and pepper noise variance = 0.02 | 20.58 | 0. 97 | 0.95 | 32.50 | 0.51 | 0.34 | salt and pepper noise variance = 0.02 |

_{1}) equals .92 and .89 for watermark2 (W_{2}). In Image 2, the average correlation for recovered W_{1} equals .91 and .88 for W_{2}. In Image 3, the average correlation for recovered W_{1} equals .92 and .89 for W_{2}. In Image 4, the average correlation for recovered W_{1} equals .93 and .88 for W_{2}.

Image | Attack type | PNSR | Correlation of watermarked 1 | Correlation of watermarked 2 |
---|---|---|---|---|

Img 1 |
Without attack | 39.06 | 0.99 | 0.97 |

Median filter [3 3] | 42.48 | 0.90 | 0.93 | |

Gaussian noise variance = 0.1 | 12.06 | 0.99 | 0.84 | |

Rotation 20 | 14.18 | 0.67 | 0.82 | |

Cropping | 17.19 | 0.98 | 0.94 | |

Salt and pepper noise variance = 0.03 | 19.42 | 0.99 | 0.88 | |

Img 2 |
Without attack | 39.16 | 0.99 | 0.96 |

Median filter [3 3] | 42.32 | 0.88 | 0.91 | |

Gaussian noise variance = 0.1 | 11.67 | 0.99 | 0.84 | |

Rotation 20 | 11.89 | 0.64 | 0.79 | |

Cropping | 12.16 | 0.97 | 0.94 | |

Salt and pepper noise variance = 0.03 | 19.99 | 0.99 | 0.89 | |

Img 3 |
Without attack | 39.16 | 0.99 | 0.96 |

Median filter [3 3] | 42.46 | 0.92 | 0.92 | |

Gaussian noise variance = 0.1 | 12.22 | 0.99 | 0.85 | |

Rotation 20 | 14.32 | 0.66 | 0.79 | |

Cropping | 17.11 | 0.97 | 0.94 | |

Salt and pepper noise variance = 0.03 | 19.12 | 0.99 | 0.88 | |

Img 4 |
Without attack | 39.06 | 0.99 | 0.97 |

Median filter [3 3] | 42.25 | 0.91 | 0.91 | |

Gaussian noise variance = 0.1 | 12.03 | 0.99 | 0.84 | |

Rotation 20 | 14.61 | 0.71 | 0.83 | |

Cropping | 19.48 | 0.99 | 0.88 | |

Salt and pepper noise variance = 0.03 | 17.34 | 0.97 | 0.89 |

Attacks | Proposed technique + encryption-additive. “correlation” | Proposed technique without encryption-additive. “correlation” | Proposed technique + encryption-LSB “correlation” | Proposed technique without encryption-LSB “correlation” | ||||
---|---|---|---|---|---|---|---|---|

Without attack | 0.97 | 0.97 | 0.97 | 0.96 | 0.95 | 0.91 | 0.95 | 0.91 |

Median filter [3 3] | 0.74 | 0.86 | 0.65 | 0.76 | 0.92 | 0.56 | 0.89 | 0.44 |

Gaussian noise variance = 0.1 | 0.95 | 0.84 | 0.79 | 0.49 | 0.42 | 0.25 | 0.38 | 0.24 |

Image rotation | 0.61 | 0.60 | 0.50 | 0.34 | 0.59 | 0.63 | 0.70 | 0.61 |

Cropping | 0.95 | 0.95 | 0.96 | 0.96 | 0.88 | 0.89 | 0.90 | 0.89 |

Salt and pepper noise variance = 0.02 | 0.97 | 0.95 | 0.31 | 0.20 | 0.51 | 0.34 | 0.41 | 0.32 |

The experimental results show that the modified watermarking technique-using additive in DWT domain with encrypted watermarks using circular and chaotic encryption to watermarks technique enhances the imperceptibility measurements PSNR as well as the robustness of the system against attacks such as data filtering. The modified technique improves robustness against all attacks than watermarking without encryption. In addition, the technique-using additive is better than using LSB.

The authors extend their appreciation to Princess Nourah bint Abdulrahman University Researchers Supporting Project Number (PNURSP2022R308), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

This work was supported by

The authors declare that they have no conflicts of interest to report regarding the present study.