Watermarking scheme based on hybrid dtcwt-dct with secant optimization method / Mohamed Lebcir

Faculty of Computing

Thesis (Doctor of Philosophy) -- Universiti Malaysia Pahang – 2025

One of the main challenges in image watermarking is achieving the optimal balance between imperceptibility and the robustness of the watermark against various types of attacks such as compression, filtering, and geometrical attacks. Researchers have been working on developing watermarking schemes that can withstand these attacks and still be able to extract the watermark without affecting the visual quality of the watermarked image. An optimized embedding strength by combining the hybrid domain with the optimization methods have succeeded in improving their performance in terms of balancing imperceptibility and robustness. However, there is a room for enhancements in achieving the aim of robustness against these types of attacks. Thus, this study proposed a scheme that combines between hybrid domain Dual Tree Complex Wavelet Transform- Discrete Cosine Transform (DTCWT-DCT) and the Secant (quasi-Newton) optimization method. The properties of the DTCWT domain can strongly deal with geometric attacks, and the properties of the DCT domain in the middle sub-band frequencies that can withstand noise and compression attacks. While the Secant method will be used to find an optimal value of the scaling factor that controls the level of watermark embedding. The main objective of this study is to produce an optimized blind DTCWT-DCT scheme that can provide a strong balance between robustness and imperceptibility. To achieve this objective, two correlated DCT-transformed sub-vectors are generated from DTCWT’s coefficients of the whole host image. Next, an objective function in the Secant method that combining Peak Signal to Noise Ratio (PSNR) and Normalized Cross-Correlation (NC) values is formulated. The function is deployed in the embedding scaling factor calculation by using only the high energy bands of the previous sub-vectors using a differential embedding manner. In doing that, the watermark extracting process also could be performed without resorting to the original host image. The scheme effectively fulfils the robustness requirement of watermarking by produced high invisibility of the watermarked image with a peak signal-to-noise ratio (PSNR) value of 45dB. It also achieved high level of strength against common image-processing attacks such as noise addition, filter, compression and geometric ones than the existing schemes. The mentioned balance requirement is also enhanced to a high degree through the optimization between the watermarked image's visual quality and the robustness of the embedded watermark. The previous characteristics ensure that the proposed scheme could be used for protecting copyrights of digital images.