DEVELOPMENT OF A ROBUST DIGITAL IMAGE WATERMARKING TECHNIQUE IN DISCRETE ORTHONORMAL STOCKWELL TRANSFORM DOMAIN BASED ON PHOTON POLARIZATION

By

Author

Presented To

Department of Engineering

ABSTRACT
This thesis presents the development of novel digital image watermarking technique for the shielding of intellectual property rights of digital images. Because of their digital nature, multimedia data can be duplicated, modified, and transformed. In this context, it is essential to develop a watermarking –based technique for copyright protection and authentication of digital images. The major problem of watermarking technique is how to achieve an optimal-tradeoff between robustness and imperceptibility. A robust digital image watermarking technique has been developed based on cascaded Discrete Orthonormal Stockwell Transform (DOST), Discrete Wavelet Transform (DWT), and Singular Value Decomposition (SVD) using private key obtained from the implementation of a Quantum Key Distribution (QKD) scheme and optimal scaling factor selection using Particle Swarm Optimization (PSO) algorithm. The colour images acquired are standard colour Lena, Pepper, Mandril, and Nepal Telecom logo. The colour images were decomposed into the respective three colour channels, of Red (R), Green (G), and Blue (B). The developed technique is achieved by applying DOST scheme on each channel of the cover and watermark images to obtain DOST transformed coefficients, these are then fed to the DWT for further transformation by decomposing the coefficients up to third level for complete representation and interpretation of the images in order to produce robust and scalable images. The images are modified by applying SVD to improve the security. A key length of 64-bits generated from the implementation of QKD scheme is used to encrypt the transformed watermark to further enhance the security of the technique. The watermark is reshuffled by affine transform by redistributing the pixel values to different locations using four 8-bit keys for it to be robust to attacks. The unintelligible transformed image by affine transform is divided into 2 pixels x 2 pixels blocks and each of the block is encrypted using XOR operation by four 8-bit keys. The optimal scaling factor obtained is used to embed and extract the watermark. The performance of the scheme is evaluated using Peak Signal to Noise Ratio (PSNR), Normalized Correlation Coefficient (NCC), and Structural Similarity Index Measure (SSIM) as performance metrics for imperceptibility and robustness. The results of the PSNR obtained for accessing the fidelity of the watermarked image when subjected to Gaussian, salt and pepper, speckle, rotated at 45°, 5°, and 90°, and crop attacks are 50.2752dB, 47.7293dB, 45.6404dB, 46.8626dB, 44.6045dB, 47.9442dB and 46.4067dB respectively, indicating that perceptual quality has been improved. The NCC values are 1, 1, 1, 0.9997, 1, 1 and 1, signifying high resistance to attacks. The SSIM values are 1, 1, 1, 0.99998, 0.9996, 0.9999 and 0.9987, respectively. The developed scheme with the QKD scheme delivered an imperceptibility improvement when compared with ITU-R standard value of 35dB for watermarking technique by 30%, 27%, 23%, 25%, 22%, 27% and 25% respectively. Comparison of the developed algorithm with the work of Bajracharya & Koju (2017) in terms of perceptual quality of the watermarked image showed an average PSNR value of 64.3580dB as against 48.1819, representing about 25% improvement of imperceptibility. The NCC values of the work of Bajracharya & Koju (2017) when subjected to the following attacks: Gaussian, salt and pepper, speckle, rotated at 45°, 5°, and 90°, and crop attacks are 1, 0.9996, 0.9999, 0.9999, 1 and 1. This demonstrates that the developed scheme gave an average improvement of about 14% robustness to attacks when compared with the work of Bajracharya & Koju (2017)