Comparative study of CNN models for detecting altered and manipulated images

Forgery in images and videos, facilitated by advanced tools like Adobe Photoshop and sophisticated algorithms, has become a significant concern in our society. The rise of AI technology has made these forgeries incredibly realistic, often indistinguishable from authentic content. Unfortunately, these deceptive media pieces are exploited in criminal activities, underscoring the urgent need for effective detection methods. Traditional techniques for identifying these forgeries struggle to keep pace with technology. To tackle this challenge, the field of multimedia forensics has begun developing various techniques, including machine learning tools, to authenticate the digital content. In our research, we compared three widely used pre-trained convolutional neural networks (CNNs): ResNet-50, LeNet-5, and AlexNet. Among these, LeNet-5 exhibited superior performance. Recognising LeNet-5's potential, we utilised it as the foundation to create and optimise two new models.