Authentication of Holograms on Video Frames Using Convolutional Neural

In today’s digital age, holograms have gained widespread recognition as a reliable indicator of authenticity on items such as credit cards, passports, and currency notes. They are also employed in consumer products like premium wines and branded merchandise to add a unique and recognizable feature. However, the proliferation of counterfeit holograms has become a growing cause for concern. As counterfeiters become increasingly sophisticated, traditional methods of hologram verification, such as visual inspection by humans, are no longer sufficient, especially when dealing with a large volume of in-product holograms. Therefore, there is a pressing need for automated tools for hologram authentication that can quickly and accurately differentiate between genuine and fake holograms.This thesis investigates the potential of convolutional neural networks (CNNs) for distinguishing between authentic and fake holograms. A deep learning framework is proposed, optimized for resource-limited devices like smartphones and utilizing a single frame input captured under ambient light. Transfer learning is employed to build upon pre-trained CNN models, reducing computational complexity. The primary objectives include developing a CNN-based classifier, designing and annotating a database of video footage, evaluating the classification system’s performance on a single frame without additional illumination, and assessing performance using accuracy, precision, recall, and F1-score metrics. Ultimately, this research aims to enhance the security and authenticity of holographic products by accurately distinguishing between authentic and fake holograms in video footage.