Investigating the Effectiveness of Deep Learning and CFA Interpolation Based Classifiers on Identifying AIGC
Michael Reidy, Henry Mallon, Jiebo Luo · 2023
AIGC is content that is generated by any AI framework. In recent years, the ability to generate AIGC has become more accessible to members of the public. This poses a threat, as AIGC has become indistinguishable to the human eye, and could be used to spread misinformation. In this paper, we investigate the ability of three deep neural networks to identify AIGC. We also propose two classifiers that use CFA Interpolation related approaches that identify if images contain genuine levels of interpolation error to identify AIGC. Our models were trained and tested on their ability to classify both previously seen and unseen image subjects to test their generalizability. It was found that CNNs performed the best at this task, but the CFA models also produced reasonable results.