BLIND DIGITAL IMAGE WATERMARKING USING SVD ANDQIM

Abstract

The proliferation of digital images over the Internet has raised concerns about copyright infringement and unauthorized reproduction of content, making robust ownershipverificationmechanismsmorenecessarythanever. Inthisthesis,afullyblind digital image watermarking scheme based on Singular Value Decomposition (SVD) and Quantization Index Modulation (QIM) in an energy-normalized reconstruction framework is proposed. A 512-bits binary watermark is embedded into singular values of 512×512 grayscale host image by using scalar QIM with quantization step ∆ = 5. The orthogonal matrices of the SVD are not altered during this process in order to keep the visual fidelity. We then perform blind extraction using the QIM minimum-distance decoder directly on the singular values of the attacked image– no reference to the original host is needed. An energy normalization mechanism is embedded to compensate for global amplitude distortions induced by energy changing attacks like histogram equalization, which stabilizes correlation based detection. The scheme was tested on four standard USC-SIPI test images under twelve attack types including noise, filtering, JPEG compression, geometric distortion and histogram modification. The imperceptibility results were near-lossless with PSNR values in the range of 68.05 to 68.44 dB an improvement of 20-28 dB over the most closely related state-of-the-art methods. Robustness was validated with Normalized Cross-Correlation values above 0.90 for all the 48 attack-image combinations, the overall mean value of NCC is 0.9845. Specifically in histogram equalization, energy normalization improved NCC by 0.47 0.52. The proposed scheme, in combination, uniquely combines the near-lossless imperceptibility, consistent robustness and the fully blind extraction to provide a practically deployable framework for the digital image copyright protection.