A method for detecting facial depth-forge created by static and dynamic clues and its characteristics

With the rapid development of deep learning technology and the swift rise of generative artificial intelligence, the quality of face forgery generation has continuously improved, and its potential risks of misuse have attracted increasing attention. This paper, which makes a systematic review of research in the related field. introduces the existing face deepfake detection methods and categorizes them according to detection cues into static detection methods and dynamic detection methods. Static detection methods include explicit logical inconsistency detection and deep feature discrepancy detection, which identify forgery traces by analyzing various differences between forged images or videos and authentic ones. In contrast, dynamic detection methods mainly focus on the temporal characteristics of videos and the consistency across different modalities. In addition, this paper reviews common face forgery techniques as well as widely used datasets for forged face images and videos, and conducts an in-depth discussion on active detection strategies and approaches for improving generalization capability.