Face counterfeit video detection method based on multi-time-sequence attention network

• Main Authors: HE ZHAOFENG, RU YIWEI, SUN ZHENAN
• Format: Patent
• Language: Chinese; English
• Published: 23.12.2022
• Subjects: CALCULATING; COMPUTING; COUNTING; PHYSICS

The invention discloses a face counterfeit video detection method based on a multi-time-sequence attention network, and the method comprises the steps: taking an attention feature map in a frame of a review sequence as feature map weight information through a review branch network, and detecting an artifact in a to-be-processed frame in the review sequence through convolution operation; detecting the coherence of a frame to be processed in a browsing sequence through a browsing branch network and convolution operation, and fusing the frame to be processed with the output of the auditing branch network in each layer through lateral connection; and finally, the full connection layer fuses output results of respective convolution processing of the browsing branch network and the inspection branch network, and finally outputs an identification result, thereby ensuring the effectiveness of face detection and greatly improving the detection efficiency. 本发明公开一种基于多时序注意力网络对人脸伪造视频检测方法，通过一个审视分支网络将从审视序列的帧中注意力特征图作为特征图权重信息