Deepfake Detection using Deep Learning: A Two-Pronged Approach with CNNs and Autoencoders

The rise of deepfakes, synthetic media generated using deep learning to create fabricated human images, undermines trust in visual content. This paper explores deepfake detection using two deep learning approaches: a convolutional neural network (CNN) and an autoencoder. The CNN detects photos as authentic or false with an accuracy of 0.83, as well as macro and weighted average precision across both classes of 0.83. The autoencoder model reconstructs input images by identifying essential features in real images. When applied to deepfake photos, larger reconstruction errors occur as a result of differences between true and modified features, allowing the model to detect anomalies indicative of deepfakes with a 57.31% accuracy. Our analysis reveals that while the CNN offers superior detection accuracy, the autoencoder excels in highlighting subtle image manipulations. The results underscore the potential of these methods for effective deepfake detection and contribute to ongoing efforts to mitigate the societal harm posed by deepfakes.