This study presents a method for authenticating visual content by detecting and localizing forged regions using channel attention convolutional blocks, with a focus on developing end-to-end channel attention networks that can successfully infer using attention-aware features from the spatial and frequency domains. The network can generate attention-aware multi-resolution features in the spatial domain by replacing channel attention blocks with basic blocks. A second attention network, in addition to the main network in the frequency domain, is used to extract additional features related to re-sampling artifacts. The features extracted from both networks are concatenated to predict the mask. The attention strategy of this network allows it to focus exclusively on the forged region, boosting the generality of the network to previously unseen manipulations.
For Codes, kindly send an email to iit.sadaf@gmail.com
Reference/Citation: I. I. Ganapathi, Syed Sadaf Ali et al. Learning to Localize Image Forgery Using End-to-End Attention Network, Neurocomputing, Elsevier, 2022.