An Image Forgery Detection and Localization via U²-Net with Multi-Scale Attention Aggregation

The increasing prevalence of image tampering in the digital age presents a critical challenge to digital media security. Although numerous deep learning algorithms, particularly those based on convolutional neural networks (CNNs), have demonstrated progress in detecting image forgery, traditional CNNs often struggle to effectively utilize subtle tampering features. To address this limitation, we propose an image tampering detection and localization method based on an improved U<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-Net architecture. This method employs a nested U-structure as its backbone network, thereby facilitating effective integration of image features across various scales. Additionally, the conventional skip connection mechanism is augmented with multi-scale attention aggregation to further enhance the network's context information fusion capability. Experimental results indicate that the proposed method outperforms existing advanced deep learning approaches in terms of performance and robustness.