Development of intelligent digital watermarking via Safe Region

Abstract

Digital watermarking has proven to curb counterfeiting and piracy of digital media in issues relating to intellectual properties, authentication, and tamper detection. However, current watennarking system is challenged by inevitable distortion caused by data embedding and these greatly affect the quality of the image and consequently reduce the specificity and completeness of the information. Furthetmore, such distortion may result in vulnerability of the algorithm to attacks, primarily, because there is no algorithm for identifying Safe Region (SR) to embed the watermark. However, artificial intelligence paradigm based on Artificial Neural Networks (ANN) is endowed with features such as detection, classification, learning and adaption suited for indentifying of SR, thus this research work presents a novel approach to improving security fidelity in watermarking system. An analysis of the significance of embedding in SR of host image in making a tradeoff between robustness and imperceptibility was carried out in this work. A hybrid of ANN and Fast Fourier Transform (FFT) algorithm was developed and implemented to determine the SR to embed watermark. It was discovered that determination of SR to embed watermark is not only important for making a superior trade off, but could also be used to tailor the watermark in accordance to an anticipated attack. Furthennore, the study showed that the unesiablished watermark embedding strength contributes to image distortion and lack of robustness during embedding. These challenges were approached with the application of ANN based embedding strength analysis for choosing embedding strength. Thus, the image distortion can be reduced efficiently, thereby improving the robustness of the watermarked image. The concept of Complex Valued Neural Network (CVNN) embedding has been proposed to implement a Damageless watermarking which prevents loss of information as well as efficient management of distortion-free images used for medical purposes. Twelve modules were developed through the use of four different algorithms; No Neural Network (No NN), Real Valued Neural Network (RVNN), CVNN and CVNN-Damageless. Host images were transformed from spatial to frequency domain using FFT to obtain the complex values. The newly proposed algorithm is blind; experiments carried out on this novel system showed it to be highly imperceptible and suitable for tamper detection applications. The results showed that embedding in the Safe Region of host image was able to fulfil the imperceptibility and robustness required for watermarking system. Embedding in SR showed Image Fidelity Measure (IFM) equal to 0.9558 of 1.0000 was obtained, signifying that there was almost no loss of fidelity in the watermarked image. The CVNN based algorithm achieved I :2.5 enhancements over the PNSR benchmark in terms of imperceptibility. The fitness of CVNN algorithm also recorded R2 of 0.92 of 1.00 implying accuracy of the model. The CVNN-Damage less results also showed no distortion between the host and the watermarked image by recording 99.99% accuracy. Thus, CVNN and CVNN Damageless proved the most efficient of existing techniques hence recommended for use in tamper detection applications, protecting integrity of sensitive images, and others images in fields which rely on authenticity of protected data.