DeePhyNet: Toward Detecting Phylogeny in Deepfakes

Deepfakes have rapidly evolved from their inception as a niche technology into a formidable tool for creating hyper-realistic manipulated content. With the ability to convincingly manipulate videos, images, and audio, deepfake technology can be used to create fake news, impersonate individuals, or even fabricate events, posing significant threats to public trust and societal stability. The technology has already been used to generate deepfakes for a number of the above-listed applications. Extending the complexities, this paper introduces the concept of deepfake phylogeny. Currently, multiple deepfake generation algorithms can also be used sequentially to create deepfakes in a phylogenetic manner. In such a scenario, deepfake detection, ingredient model signature detection, and phylogeny sequence detection performances have to be optimized. To address the challenge of detecting such deepfakes, we propose DeePhyNet, which performs three tasks: it first differentiates between real and fake content; it next determines the signature of the generative algorithm used for deepfake creation to determine which algorithm has been used for generation, and finally, it also predicts the phylogeny of algorithms used for generation. To the best of our knowledge, this is the first algorithm that performs all three tasks together for deepfake media analysis. Another contribution of this research is the DeePhyV2 database to incorporate multiple deepfake generation algorithms including recently proposed diffusion models and longer phylogenetic sequences. It consists of 8960 deepfake videos generated using four different generation techniques. The results on multiple protocols and comparisons with state-of-the-art algorithms demonstrate that the proposed algorithm yields the highest overall classification results across all three tasks.