Research on smartphone image source identification based on PRNU features collected multivariate sampling strategy

Abstract

Photo Response Non-Uniformity (PRNU)-based image source attribution is one of the core methods for identifying the imaging device of a given picture, and has significant applications in the field of digital media forensics. However, with the increasing complexity of smartphone imaging systems, PRNU features extracted from smartphone images exhibit greater instability compared to those from traditional cameras. This instability can lead to performance degradation in conventional single-sample extraction strategies when applied to smartphone image source attribution. To address this challenge, this paper proposes a robust multi-sample enhancement scheme. To verify its generalizability, we employ both a non–data-driven wavelet-domain decomposition algorithm and a deep U-shaped residual neural network (DRUNet) as noise extractors, and conduct experiments on the FODB dataset. Experimental results demonstrate that the proposed multi-sample framework exhibits superior performance in improving feature stability, providing a new technical pathway for digital image source attribution in smart terminal devices. Furthermore, we perform PCE distribution statistics on positive and negative samples in the dataset and quantitatively analyze the regional instability of PRNU features.