Detection of Leading CNN Models for AI Image Accuracy and Efficiency
Meshal Nayim, Vishnu Mohan, Trilok Nath Pandey, Bibhuti Bhusan Dash, Bibek Bikram Dash, Sudhansu Shekhar Patra · 2024
This research explores the challenge of distinguishing between authentic and AI-generated images using advanced convolutional neural network (CNN) architectures. Leveraging the CIFAKE dataset, which includes a diverse array of AI-generated and real images, we evaluated the performance of three pre-trained models: ResNet50, EfficientNetB0, and DenseNet121. Our study involved training these models with images of 32x32 and resized 64x64 pixel resolutions to assess their classification accuracy, precision, recall, and computational efficiency. DenseNet121 emerged as the most effective model, achieving an accuracy of 98.49% on the test dataset, significantly outperforming the other models. This work highlights the potential of deep learning methodologies in addressing the growing need for reliable AI image detection, providing a robust foundation for future advancements and applications in this critical field.