Digital Media Watermark Attack Algorithm Based on Multi-Scale Feature Fusion

To enhance the accuracy of watermark attacks and preserve image fidelity in digital media products, this study proposes MSAFAN, a watermark attack algorithm based on multi-scale feature fusion. The method strengthens multi-scale perception through feature decomposition and dynamic gating mechanisms. Experiments conducted on the COCO2017 dataset demonstrate that MSAFAN achieves a favorable balance between image fidelity and watermark disruption under various noise conditions, outperforming traditional attack methods in overall performance. Ablation studies further confirm the critical roles of multi-scale enhancement and contextual modeling modules. This research enriches the theoretical framework of adversarial image security and provides new technical support for efficient watermark attacks.