Robust Attack Identification Strategy to Prevent Document Image Forgeries by using Enhanced Learning Methodology
R. Thandaiah Prabu, S. Diwakaran, Advait Balaji, Ch. Prathima, K M Dhanalakshmi, Sethu Vijayakumar
Abstract
Forgery in images is a long-standing issue that spans back to ancient times. Throughout history, images have been employed as evidence or documentation of events. In the present day, the accessibility of image preprocessing tools and affordable hardware has facilitated the creation of manipulated images with ease, often driven by ulterior motives such as disseminating false information or pursuing personal gains unlawfully. In recent years, the spotlight has turned to Artificial Neural Networks (ANNs), sparking significant interest in the realm of image forgery identification. Nevertheless, prevailing image forgery techniques rooted in ANNs are often specialized, geared towards identifying distinct types of forgeries. Consequently, a pressing need arises for an approach adept at efficiently and accurately uncovering unfamiliar forms of image manipulation. This research fills that gap by presenting a robust deep learning-based system designed specifically to spot fakes while working with double image compression. The model's training hinges upon discerning disparities between an original image and its recompressed iterations. To enhance the model's performance, the Ant Lion Optimization (ALO) technique is employed to fine-tune the features of the input data. Noteworthy for its efficiency, the proposed model stands out as a lightweight solution. Empirical results underscore its prowess, showcasing swifter effectiveness compared to prevailing state-of-the-art technique. Encouragingly, the model achieves an impressive overall validation accuracy of 95.87%, reinforcing its efficacy in identifying image forgeries.