Cvpr2022 𞓜 feature decoupling learning and dynamic fusion for re captured images

In this paper, the author proposes a new feature decoupling and dynamic fusion （FDDF） Modules are most effective for adaptive learning recapture features , In addition, the author has collected and produced a large-scale real-world scene recapture（RUR） Data sets , It contains various recapture Pattern , It is about five times the number of previously published data sets .

The author is the first to propose a general model and a general real scene large-scale data set for recaptured image forensic People who , What is mentioned is FDDF Can be in RUR Reached on dataset SOTA The level of .

Thesis link ：

https://arxiv.org/pdf/2206.06103.pdf

FDDF The architecture of the model , As shown in the figure below .

The model is decoupled by explicit features （EFD） Module and dynamic feature fusion （DFF） Modules are combined . The whole process can be described as ：

（1） The image is fed into four parallel branches , To unify Resnet18 Four types of backbone network generation capture feature;

（2）capture feature Fused by learnable modules with adaptive weights ;

（3） The fused features will be sent to the classification layer , To determine whether the image is recapture.