Multi branch deepfake detection based on double attention mechanism

With the continuous development of artificial face technology, it is more and more difficult to distinguish true and false faces with naked eyes. This paper proposes a multi branch artifact detection algorithm based on double attention mechanism, which can detect subtle artifacts. Firstly, dlib is used for face detection, and the local images of eyes, nose and mouth are segmented. A multi branch detection network model based on double attention mechanism is proposed. The model fully learns the context semantic information of local artifacts and global artifacts. The performance of the proposed method is evaluated on the deepfake data set, and the experimental results show that the test accuracy of the proposed method reaches 96.45%. At the same time, compared with other methods, this method also has a good performance in anti compression test.