Copy-move detection method based on Decoupled Edge Supervision and multi-domain cross correlation modeling
Niantai Jing,Jie Nie,Jingyu Wang,Xiaodong Wang,Xinyue Liang,Xuesong Gao
DOI: https://doi.org/10.1007/s11042-024-19584-z
IF: 2.577
2024-07-20
Multimedia Tools and Applications
Abstract:Copy move forgery detection is defined as the process of moving one region in the image (the source region) to another region in the image (the tampered region). Common classification methods include copy move forgery location (CMFL, which does not distinguish between source regions and tampered regions) while existing copy move source target detection (CMSTD) use limited information to distinguish the source and region. Since the edges of the tampered region able to act as the important clues such as blurring, bringing the challenge of detecting the copy-move forgery target with the complete structure. Therefore, we propose a Copy-move Detection Method based on Decoupled Edge Supervision and Multi-domain Cross Correlation (DM-Net), including Multi-scale Similar Region Detection module (MSD), Decoupled Edge Supervision module (DEM), and Multi-domain Correlation Modeling module (MCM), which can overcome the problem that the tamper trace is fuzzy caused by the post-processing operation. Specifically, the MSD module is proposed to extract coarse similar regions by multi-scale method. The DEM module is proposed to extract the tamper region by the method of decoupling edge supervision, which avoids information redundancy while using shallow edge features. The MCM module conducts cross-correlation modeling between the tampered, source and similar region, further optimizes detection targets of similar region by mining the correlation among multiple domains. By adding edge information, we can improve the efficiency of distinguishing source and target regions by 2%. We performed experiments on USC-ISI data set, and the accuracy was improved by 0.21% compared with CNN-T GAN method, and the F1-score index was improved by 0.87% compared with DOA-GAN. The accuracy of CASIA v2.0 data set is 2.89% higher than that of Busternet method, and the precision index is 3.98% higher than that of CMSD-STRD method on source. The accuracy of CoMoFoD data set is improved by 0.93% compared with CMSD-STRD method, and the recall index is improved by 20.23% compared with CMSD-STRD target, which proves that our method is superior to the most advanced method at present.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering