AI vs. Human Vision: A Comparative Analysis for Distinguishing AI-Generated and Natural Images

Today’s data-driven generation has led to remarkable advancements in technology. However, as there are two sides to a coin, technology too has both its advantages and disadvantages. The expansion of AI has given rise to ‘Deepfake’ which involves skillful superimposing of person’s face with another person’s face which is very dangerous and it is used to produce morphed images and disseminate fake videos which has led to cyberbullying, financial fraud and cybersecurity risks. Our goal is to correctly determine authentic images by classifying them into AI generated v/s real images.We have used ‘PyGoogle’ image library for creation of dataset for AI images and for the real image dataset we have used our own camera to capture real images. We have used CNN model on both the dataset and observed that accuracy of Google images dataset is 88 percent and that of the own dataset is 81 percent. For evaluating the performance of our model we have created Confusion Matrix for the same.