Image Upsampling Detection Based on Autocorrelation Matrix
Feifan Wu, Xiaolong Li, Jingtian Wang, Yao Zhao · 2024
Upsampling detection is a key issue of digital image forensics. Among the widely used methods, the one based on spectrum analysis suffers from the aliasing problem, necessitating a prior constraint on the estimation range of interpolation factors. To address this limitation, this paper introduces a novel second-order metric and proposes an analysis approach based on the autocorrelation matrix. The study reveals that the Fourier spectrum of the autocorrelation matrix of upsampled images contains detectable features at specific positions, which can effectively eliminate aliasing effects. Building upon this, a blind and efficient method for detecting up-sampling traces is proposed. Extensive experiments demonstrate that the proposed detector outperforms state-of-the-art methods.