Copy-Move Detection in Optical Microscopy: A Segmentation Network and A Dataset

TL;DR

This work tackles the detection of copy-move forgeries in optical microscopy by reframing the problem as intra-image co-saliency detection and proposing CMSeg-Net, a lightweight segmentation network enhanced with a correlation-assisted spatial-attention (CoSA) mechanism. The model integrates a MobileNet-v2 encoder, a multi-resolution decoder with inverted residual blocks, and a VRSA-based attention strategy to capture intra-image region correlations across scales, enabling detection of small or background-like duplicates and unseen attacks. A new dataset, FakeParaEgg, simulates realistic copy-move manipulations in microscopic images and, together with evaluation on CASIA-CMFD, CoMoFoD, and CMF, demonstrates CMSeg-Net's superior performance and robust generalization to out-of-domain data. Overall, the approach offers a practical, efficient tool for forensic analysis of biomedical images and provides a new benchmark to spur further research in this niche domain.

Abstract

With increasing revelations of academic fraud, detecting forged experimental images in the biomedical field has become a public concern. The challenge lies in the fact that copy-move targets can include background tissue, small foreground objects, or both, which may be out of the training domain and subject to unseen attacks, rendering standard object-detection-based approaches less effective. To address this, we reformulate the problem of detecting biomedical copy-move forgery regions as an intra-image co-saliency detection task and propose CMSeg-Net, a copy-move forgery segmentation network capable of identifying unseen duplicated areas. Built on a multi-resolution encoder-decoder architecture, CMSeg-Net incorporates self-correlation and correlation-assisted spatial-attention modules to detect intra-image regional similarities within feature tensors at each observation scale. This design helps distinguish even small copy-move targets in complex microscopic images from other similar objects. Furthermore, we created a copy-move forgery dataset of optical microscopic images, named FakeParaEgg, using open data from the ICIP 2022 Challenge to support CMSeg-Net's development and verify its performance. Extensive experiments demonstrate that our approach outperforms previous state-of-the-art methods on the FakeParaEgg dataset and other open copy-move detection datasets, including CASIA-CMFD, CoMoFoD, and CMF. The FakeParaEgg dataset, our source code, and the CMF dataset with our manually defined segmentation ground truths available at ``https://github.com/YoursEver/FakeParaEgg''.

Figures (3)

• Figure 1: Architecture of the CMSeg-Net. The encoding path from $\mathbf{T}_1$ to $\mathbf{T}_6$ follows MobileNetV2, using ReLU6 activation for robustness, while the decoding path from $\mathbf{R}_5$ to $\mathbf{R}1$ utilizes the inverted residual block (IRB) for reconstruction. The VRSA block enhances its input with a spatial attention submodule featuring a variable-sized receptive field, while the CoR block derives spatial correlation for copy-move detection. Note that $C_2=16$, $C_3=24$, $C_4=32$, $C_5=96$, $C_6=1280$, and $(h_{i+1}, w_{i+1})=(\frac{1}{2}h_i,\frac{1}{2}w_i)$.
• Figure 2: Visual performance comparison among methods on FakeParaEgg. Note that Columns iv-vii are out-of-domain generalizability tests.
• Figure 3: Visual performance comparison of our method against ManTra-Net, DOA-GAN, and UCM-Net on the CASIA-CMFD dataset.