基于面部动态特征和注意力机制的生成人脸检测OA

The development of AI,machine learning and deep learning has given birth to a large number of multimedia processing technologies and tools.Although they are mostly used in legal fields such as entertainment and education,malicious abuse has caused concerns,and face generation technology is typical.The existing generative face detection method focuses on visual appearance features,and has problems such as complex models,susceptible to noise,and low detection accuracy.In this regard,we propose a new generative face detection framework.This framework effectively captures abnormal changes in facial landmark points by analyzing the changes in dynamic characteristics of face landmark points in a video sequence,and combining it with timing modeling and attention mechanisms,thereby improving the accuracy of generating face detection frameworks.Experimental evaluations on multiple data sets prove the effectiveness of the proposed method,where the accuracy rate of the proposed method is as high as 99.76%on the FaceForensics++dataset,demonstrating its superior performance in detecting and generating faces.At the same time,when the proposed method processes c23(light compression)and c40(severe compression),the accuracy rate decreased by only 2.63 percentage points and 10.16 percentage points respectively,effectively verifying its robustness for video compression.