Video inter-frame forgery detection based on CNN and ViT
Nan Zhu, Xin Wang, Kun Wang, Xu Zhang · 2025
With the explosive advancements in user-friendly video altering tools, digital video forgery becomes very convenient and easy, which makes the verification of video authenticity significant. Among various video forgeries, video interframe tampering operations, which include frame deletion, insertion, and duplication, are the most commonly-used operations. In order to prevent such a security loophole, we propose a video inter-frame forgery detection method based on CNN (convolutional neural networks) and ViT (vision Transformer). Specifically, we first calculate the pixel-wise difference of adjacent frames and split the obtained differential sequence into overlapping clips. Then a modified pretrained VGG16 is utilized to extract features from each frame in clip. Finally, the output of a series of Transformer encoders is utilized to make final decision. Extensive experiments on the public VIFFD and TDTVD datasets demonstrate the superiority of our proposed method on detection accuracy.