Research on identity document image tampering detection based on texture understanding and multistream networks
Lixin Wang, Zhenjiang Li, Wufan Zhao · Journal of Electronic Imaging · 2025
Identity document images are widely utilized in personal identity authentication systems. However, the rapid development of image-editing tools poses significant risks to personal privacy and information security. Consequently, tampering detection in identity document images has become a crucial research focus. The current challenges in this field primarily stem from two aspects: the increasing realism of tampered content and the prevalence of secondary editing operations on manipulated images. We propose a multistream network architecture that leverages the distinctive background texture patterns of identity documents. This model enables simultaneous extraction and effective fusion of three critical feature domains: background texture characteristics, frequency-domain representations, and noise distribution. To support rigorous evaluation, we specifically develop the tampered-identity dataset containing primarily secondarily edited tampered samples. Experiments conducted on the tampered-identity dataset as well as other publicly available datasets demonstrate that the proposed method exhibits commendable detection performance, particularly in effectively identifying tampering that has been formed through secondary editing.