Detection of Artificially Generated Synthetic Images Using Parallel Convolutional Neural Network

The generation of extremely realistic AI generated synthetic images which are extremely difficult to differentiate from real ones has been made possible by the fast development of new AI generative models, in particular diffusion models. With the purpose of precisely detecting these artificially generated synthetic images, this study suggests a novel detection method that uses a Parallel Convolutional Neural Network. This new design improves the model's ability to identify minute discrepancies introduced during image production by including two concurrent convolutional branches, one for capturing global features and the other for capturing local features. The proposed model gives an accuracy of 94.7% and uses Gradient Class Activation Mapping to implement Explainable AI to visualize the model's decision making.