Forensic analysis of manipulated pixels requires the identification of
various hidden and subtle features from images. Conventional image recognition
models generally fail at this task because they are biased and more attentive
toward the dominant local and spatial features. In this paper, we propose a
novel Gated Context Attention Network (GCA-Net) that utilizes non-local
attention in conjunction with a gating mechanism in order to capture the finer
image discrepancies and better identify forged regions. The proposed framework
uses high dimensional embeddings to filter and aggregate the relevant context
from coarse feature maps at various stages of the decoding process. This
improves the network's understanding of global differences and reduces
false-positive localizations. Our evaluation on standard image forensic
benchmarks shows that GCA-Net can both compete against and improve over
state-of-the-art networks by an average of 4.7% AUC. Additional ablation
studies also demonstrate the method's robustness against attributions and
resilience to false-positive predictions.

本论文提出一种新颖的门控上下文注意网络（GCA-Net），使用非局部注意机制和门控机制捕捉图像不一致特征，以更好地识别图像中伪造的区域。经过标准图像取证测试，该方法相对于最先进的网络平均值提高了 4.7% 的 AUC，并且对归因具有鲁棒性，并且对误报具有抗性。