| 電子情報通信学会 研究会発表申込システム 講演論文 詳細 |
技報閲覧サービス [ログイン] 技報アーカイブ |
| トップに戻る | 前のページに戻る | [Japanese] / [English] |
| 講演抄録／キーワード | ||
| 講演名 | 2025-11-04 09:50 Network-Based AI Recognition of Authentic and Forged Images Jiann-Liang Chen・○Chia-Ying Lin・Hau-Ching Chen・Ren-Bao Zhou・Yi-Hong Yeh・Yan-Ting Pan・Yan-Xin Tsai・Shih-Ping Chiu（NTUST）  IA2025-36 |
|
| 抄録 | （和） | To mitigate the spread of misinformation caused by AI-generated media online, we propose a novel detection model that leverages the pre-trained CLIP image encoder. Our approach systematically selects an efficient CLIP variant and then feeds its powerful semantic features into a Transformer classifier head designed to distinguish authentic images from forged ones. Evaluated on the GenImage dataset, our model achieves a mean cross-validation AUC of 96.25% across all subsets, surpassing previous CLIP-based methods. This work demonstrates that coupling large-scale vision models with a simple Transformer head is an effective and computationally efficient strategy for image authenticity verification. |
| （英） | To mitigate the spread of misinformation caused by AI-generated media online, we propose a novel detection model that leverages the pre-trained CLIP image encoder. Our approach systematically selects an efficient CLIP variant and then feeds its powerful semantic features into a Transformer classifier head designed to distinguish authentic images from forged ones. Evaluated on the GenImage dataset, our model achieves a mean cross-validation AUC of 96.25% across all subsets, surpassing previous CLIP-based methods. This work demonstrates that coupling large-scale vision models with a simple Transformer head is an effective and computationally efficient strategy for image authenticity verification. | |
| キーワード | （和） | AI-generated media / CLIP / Transformer classifier / image authenticity verification / / / / |
| （英） | AI-generated media / CLIP / Transformer classifier / image authenticity verification / / / / | |
| 文献情報 | 信学技報, vol. 125, no. 222, IA2025-36, pp. 3-8, 2025年11月. | |
| 資料番号 | IA2025-36 | |
| 発行日 | 2025-10-28 (IA) | |
| ISSN | Online edition: ISSN 2432-6380 | |
| 著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します．(許諾番号：10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) | |
| PDFダウンロード | IA2025-36 |
|
[研究会発表申込システムのトップページに戻る]