Art of Detection: Custom CNN and VGG19 for Accurate Real Vs Fake Image Identification
V Hayathunnisa, Kuppusamy P, A. Manimaran · 2023
In the current digital age, the ability to distinguish between real and manipulated images has become crucial due to the proliferation of doctored images. This research aims to address this challenge by employing two distinct neural network architectures: a custom Convolutional Neural Network (CNN) and the pre-trained VGG19 model for the classification of real versus fake images. Experimental results reveal that the custom CNN achieved a noteworthy accuracy of 94.46%, outperforming the VGG19 model, which secured an accuracy of 84.24%. Such findings suggest that while pre-trained models like VGG19 bring significant value to image classification tasks, a tailored CNN can offer superior performance for specialized tasks such as detecting image authenticity. This study provides a foundation for further exploration in image forensics, emphasizing the importance of model selection and optimization in combating digital image manipulations.