Siameformer: Towards a Robust Source Camera Identification Method on Lossy Images

Abstract

Source camera identification (SCI) is a common problem in digital forensics, aiming to identify the source device of given images. Traditional digital image source forensics are primarily conducted in original scenarios. However, in real-world scenarios, digital images disseminated through social networks are often subjected to various complex interferences, leading to a significant decline in the performance of existing image source forensic methods. To address the issue of network image source forensics in real-world scenarios, we propose a network image source forensics method called Siameformer. This method integrates a multi-head attention-based Siamese network with convolutional neural networks (CNN) and Vision Transformers (ViT) for image source identification. We designed comprehensive experiments, and our model achieved satisfactory performance on benchmark public databases (such as Dresden, VISION, and Forchheim).

Keywords: Source camera identification, Multi-head attention, Siamese Network, vision transformer