Face Deepfakes -- A Comprehensive Review

TL;DR

Face Deepfakes -- A Comprehensive Review surveys state-of-the-art face deepfake generation and detection, emphasizing algorithmic details, training paradigms, losses, and evaluation metrics. It analyzes the implications for biometric recognition, catalogues positive and negative applications, and highlights gaps in robustness and regulation. Key contributions include a systematic taxonomy of generation and detection methods, a biometric vulnerability assessment, and proposed directions such as universal detectors and explainable detection. The work aims to inform researchers, policymakers, and the public about risks, mitigation strategies, and responsible pathways for ongoing deepfake research.

Abstract

In recent years, remarkable advancements in deep-fake generation technology have led to unprecedented leaps in its realism and capabilities. Despite these advances, we observe a notable lack of structured and deep analysis deepfake technology. The principal aim of this survey is to contribute a thorough theoretical analysis of state-of-the-art face deepfake generation and detection methods. Furthermore, we provide a coherent and systematic evaluation of the implications of deepfakes on face biometric recognition approaches. In addition, we outline key applications of face deepfake technology, elucidating both positive and negative applications of the technology, provide a detailed discussion regarding the gaps in existing research, and propose key research directions for further investigation.

Figures (9)

• Figure 1: The Organization of this Survey
• Figure 2: Illustration of different face generation techniques within deepfakes. Sub-figures (i)-(iv) have been sourced from , , , and , respectively.
• Figure 3: A comparison between face swapping and face reenactment processes
• Figure 4: Illustration of the autoencoder-based framework introduced for face swapping. A shared encoder generates latent representations for source and target faces and the two decoder networks recreate the inputs of their respective identities. In the face-swapping stage, the decoder of the source face is used to reconstruct the source face on the target video.
• Figure 5: Illustration of the architecture of FSNet model natsume2019fsnet which is composed of VAE-based encoder-decoder architecture, and a GAN based generator network.