Transformada Wavelet Haar discreta aproximada para o sistema de marca d'Água com eficiência energética

Abstract

This dissertation proposes the implementation of digital watermarking systems as an effective strategy to ensure the authenticity and integrity of images, especially in the context of the growing use of social media platforms. With the popularization of these platforms, the misuse of personal images and the spread of fake news have become increasingly frequent problems. Criminals have exploited the limitations of the Human Visual System (HVS) to subtly alter digital images, compromising their veracity. In this scenario, digital watermarking emerges as a promising solution, allowing the insertion of information into images in a way that is invisible to the HVS, such as owner identification, using sophisticated transform-based techniques. This work focuses on the investigation of the Wavelet Transform family, with a specific focus on the Haar, Daubechies, Coiflet, Symlet, Biorthogonal, and Reverse Biorthogonal variants for the insertion and extraction of watermarks. The research involved a detailed analysis of how these different transforms perform under a variety of common attacks described in the literature, such as Gaussian noise, JPEG compression, median filtering, and salt-and-pepper noise. The analyses indicated that the Haar, Biorthogonal, and Reverse Biorthogonal transforms stand out for their robustness and effectiveness in watermark insertion and extraction, making them the most suitable for developing image protection systems. Based on these findings, an innovative hardware watermarking system was developed using the Haar Wavelet Transform in combination with approximate computing techniques. The primary focus of this development was to meet the needs of applications demanding low energy consump tion without compromising the integrity and security of the watermarks. The system’s implementation resulted in significant hardware optimization, with the reduction of ten arithmetic circuits, leading to a decrease in area by 1.33 times and a reduction in energy consumption by 2.88 times compared to state-of-the-art solutions.