DSTNet: Distinguishing Source and Target Areas for Image Copy-Move Forgery Detection

Kaiqi Zhao, Xiaochen Yuan, Guoheng Huang, Kun Liu

研究成果: Conference contribution同行評審

摘要

In copy-move forgery detection, most relevant studies concern locating the copy-move areas without the distinction of source and target regions. This paper proposes an end-to-end network, DSTNet, to identify the source and target based on consistency detection between the copy-move region and the non-copy-move region. The DSTNet is composed of two stages, the Pre-processing stage and the Discrimination stage. Pre-processing Stage extracts internal information of copy-move and non-copy-move areas and conducts a series of operations to meet the requirements of network input. Discrimination stage allows multiple patches for input and classifies the input patches. Specifically, the Pre-processing stage, contains the Copy-move Patches Selection (CM Patches Selection) and Genuine Patches Selection, can select pairs of copy-move and none copy-move patches. We train the proposed DSTNet on two large synthetic datasets and use the public datasets CASIA and Comofod for evaluation. The experiment shows that our method achieves excellent results. Particularly, we achieve a 5.4% higher F1 based on ground-truth of copy-move mask (GT-CM) on CASIA dataset.

原文English
主出版物標題Pattern Recognition - 27th International Conference, ICPR 2024, Proceedings
編輯Apostolos Antonacopoulos, Subhasis Chaudhuri, Rama Chellappa, Cheng-Lin Liu, Saumik Bhattacharya, Umapada Pal
發行者Springer Science and Business Media Deutschland GmbH
頁面321-333
頁數13
ISBN(列印)9783031783111
DOIs
出版狀態Published - 2025
事件27th International Conference on Pattern Recognition, ICPR 2024 - Kolkata, India
持續時間: 1 12月 20245 12月 2024

出版系列

名字Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
15322 LNCS
ISSN(列印)0302-9743
ISSN(電子)1611-3349

Conference

Conference27th International Conference on Pattern Recognition, ICPR 2024
國家/地區India
城市Kolkata
期間1/12/245/12/24

指紋

深入研究「DSTNet: Distinguishing Source and Target Areas for Image Copy-Move Forgery Detection」主題。共同形成了獨特的指紋。

引用此