Enhanced Source Camera Identification Using Dual Pathway Processing and Spatial Attention Module

Source camera identification is a critical task in digital image forensics that helps verify the authenticity of images by identifying the camera sensor used to capture them. In this paper, we propose an enhanced method for source camera identification. Building upon an existing work, we introduce a model that learns to compare the camera fingerprint Photo Response Non-Uniformity (PRNU) with image noise at the patch level. First, we utilize a dual-pathway structure to process image noise and fingerprint inputs independently, allowing the network to capture specialized features from each input. Additionally, we use 3-channel inputs to preserve all the relevant noise information across all channels. This multichannel approach is combined with a spatial attention module to emphasize regions rich in PRNU information while ignoring irrelevant noise. Furthermore, experiments conducted on the VISION dataset, which includes images processed by social media platforms, demonstrate that the proposed model outperforms the baseline model in term of accuracy and robustness while maintaining computational efficiency. These improvements make our model well-suited for large-scale forensic applications in real-world scenarios.