Deep Fake Image Detection Based on Decoupled Dynamic Convolution
Jiale Li, Bicheng Li · 2022
At present, deep forge image detection algorithm has some problems such as unknowable content, insufficient generalization ability and heavy computation when detecting samples of multi-scene and multi-forgery techniques. To solve these problems, a depth forged image detection method based on decoupling dynamic convolution is proposed. Firstly, the random brightness contrast was set, and the abnormal information of the enhanced image was used as the input of the model. Then, the decoupling dynamic convolution is introduced into the residual network, and the abnormal features of the forged image are captured by channel dynamic convolution. The location of abnormal features is extracted by spatial dynamic convolution, and the effective abnormal features are extracted by synthesizing channel information and spatial information of the image to reduce the calculation cost. Finally, the full connection layer is used to detect the depth forged image. The experimental results show that the new method can distinguish the true and false images efficiently when the data sample scenes are diverse and the forging methods are abundant. The average accuracy of the new method is better than that of the comparison method, and the calculation cost is lower than that of the comparison method.