Deepfakes Detection Based on Multi Scale Fusion

Generative adversarial networks (GANs) and deep learning technologies pose a great threat to public security. The traditional forgery and tampering detectionmethods are difficult to use for the detection of such images or videos. In this paper, based on the deep learning method, the deep neural network is used to extract, fuse and classify the higher dimensional space-time features of the input image and sequence frame in the spatial and temporal dimensions, and a kind of automatic deepfakes detection technology based on multi-dimensional space-time information fusion is proposed. In the spatial dimension, using the spatial learning ability of convolutional neural network (CNN), the feature pyramid (FPN) is fused to the feature map extracted by the backbone feature extraction network for up sampling and weighted fusion. According to the results of higher dimensional feature fusion classification, deepfakes are detected.