DETECTION OF DEEP FAKE VIDEO USING LONG DISTANCE ATTENTION

Abstract

With the rapid progress of deepfake techniques in recent years, facial video forgery 
can generate highly deceptive video contents and bring severe security threats. And 
detection of such forgery videos is much more urgent and challenging. Most existing 
detection methods treat the problem as a vanilla binary classification problem. In this 
paper, the problem is treated as a special fine-grained classification problem since the 
differences between fake and real faces are very subtle. It is observed that most 
existing face forgery methods left some common artifacts in the spatial domain and 
time domain, including generative defects in the spatial domain and inter-frame 
inconsistencies in the time domain. And a spatial-temporal model is proposed which 
has two components for capturing spatial and temporal forgery traces in global 
perspective respectively. The two components are designed using a novel long 
distance attention mechanism. The one component of the spatial domain is used to 
capture artifacts in a single frame, and the other component of the time domain is used 
to capture artifacts in consecutive frames. They generate attention maps in the form of 
patches. The attention method has a broader vision which contributes to better 
assembling global information and extracting local statistic information. Finally, the 
attention maps are used to guide the network to focus on pivotal parts of the face, just 
like other fine-grained classification methods. The experimental results on different 
public datasets demonstrate that the proposed method achieves the state-of the- art 
performance, and the proposed long distance attention method can effectively capture 
pivotal parts for face forgery.