Development of Efficient Watermarking Techniques in Medical Images

Abstract

Watermarking techniques are widely used for copyright protection, confidentiality and integrity issues in medical field. There are various watermarking techniques for hiding crucial patient’s data while digital medical image transmission but most lack resistance against many unwanted attacks. There is a greater need for prevention of unauthorized access and tampering of medical may lead to misdiagnosis & wrongful treatment and can also influence the life of a human being. Reversibility, robustness, embedding capacity and invisibility are the essential requirements of a watermarking technique. This research proposes four different robust and reversible watermarking techniques for medical image watermarking. Cogitating the need of security for medical images, this first technique, proposes a reversible high embedding capacity, high image fidelity, a hybrid robust lossless data hiding technique by using both transform and spatial domains. Proposed technique alters the mean of the selected non-overlapping slantlet transformed blocks of the host image whereas RS vector considers flipping factor for data embedding. The optimum thresholds to select the blocks are calculated through Particle Swarm Optimization (PSO) technique and watermark is generated by using patient details, biometric Identification (ID) and Region Of Interest (ROI) blocks of host image. This watermark is further compressed by applying Lempel-Ziv-Welch (LZW) technique and encrypted by Advanced Encryption Standard (AES) as well as Message Digest Algorithm 5 (MD5). The watermark bits are embedded in all three Red Green Blue (RGB) channels of a cover image, to increase the embedding capacity up to 3.3675 Bit Per Pixel (bpp). The second proposed technique is a hybrid robust lossless data hiding algorithm that uses the Singular Value Decomposition (SVD) with Fast Walsh Transform (FWT) and Slantlet Transform (SLT) for image authentication. These transforms have good energy compaction with distinct filtering, which leads to higher embedding capacity from 1.8 bit per pixel (bpp) up to 7.5bpp. In this technique, Artificial Neural Network (ANN) is applied for ROI detection, and two different watermarks are created. Embedding is done after applying FWH by changing the SVD coefficients and by changing the highest coefficients of SLT subbands. In dual hybrid embedding, the first watermark is the ROI, and the other watermark consists of three parts: patients’ personal details, unique biometric ID, and the key for encryption. The third proposed technique focuses on robustness, reversible data hiding with tamper localization, and recovery for medical images. In this technique, Artificial Neural Network (ANN) is used to create a watermark creation by feature extraction, and the medical image is divided into ROI and Region of Non-Interest (RONI). Then, hybrid watermarking uses SLT and RS Vector. Secure Hash Algorithm 3 (SHA-3), AES, and LZW are used for reliability and confidentiality. Pre-processing is done to reduce the disordered pixels for minimal visual distortion and contrast enhancement. The tampered data recovery of the ROI from the watermarked image at the receiver’s end is located by the difference matrix between the ROI bits after applying ANN and the bits extracted from the background. The fourth proposed technique is called the Austere Viable Watermarking (AVW) technique. Low energy coefficients of Ridgelet Transform (RT) are used for hiding the watermark bits. AVW unravels three divergent watermarking algorithms for three different usage. Initially, a Unique Identity Document-AVW (UID-AVW) is used for mapping unique patient’s information and biometric identification as a watermark with the ROI. Then, ROI-AVW is used for embedding in the RONI by v modifying the RT coefficient’s mean values. Finally, fusion is done with the ratio of mean and variance with respect to threshold through particle swarm optimization for Tamper Detection/Recovery AVW (TDR-AVW). The credibility of the proposed techniques in comparison with other medical watermarking techniques is evidenced through experimental results. Experiments are simulated on the proposed techniques by casting numerous attacks for testing the visibility, robustness, security, authenticity, integrity and reversibility. The resultant outcome proves that the watermarked image has an improved imperceptibility with a high level of payload, low time complexity and high Peak Signal to Noise Ratio (PSNR) against the existing approaches. Experimental results demonstrate that in comparison with more than 30 existing articles, the third proposed technique achieves high robustness against more than 20 attacks along with tamper detection and recovery of ROI, preserving the visual quality of the cover image. Fourth proposed technique relinquishes much improved outcome in comparison to more than 30 published techniques in terms of attacks withholding resistance, Normal correlation, Standard deviation error, capacity, PSNR, BPP, Structural Similarity Index (SSIM) and execution time.