Meta-Learning For Image Forgery Detection: Tackling Class Imbalance With Focal Loss And Oversampling

International Journal of Electronics and Communication Engineering

© 2025 by SSRG - IJECE Journal

Volume 12 Issue 7

Year of Publication : 2025

Authors : M. Samel, A. Mallikarjuna Reddy

10.14445/23488549/IJECE-V12I7P127

How to Cite?

M. Samel, A. Mallikarjuna Reddy, "Meta-Learning For Image Forgery Detection: Tackling Class Imbalance With Focal Loss And Oversampling," SSRG International Journal of Electronics and Communication Engineering, vol. 12,  no. 7, pp. 346-359, 2025. Crossref, https://doi.org/10.14445/23488549/IJECE-V12I7P127

Abstract:

In recent years, the rapid dissemination of images manipulated in digital platforms has faced serious challenges for visual forensics, journalism, and legal integrity. To remove traditional identification techniques under the terms of image, copy-movie, and AI-acted, especially under the terms of data imbalance and limited supervision. This research introduces a novel forgery detection structure, which takes advantage of meta-learning, especially Model-Agnostic Meta-Learning (MAML), to enable rapid adaptation to the new manipulation pattern using only a small set of labelled examples. The model is trained on a balanced suite of tasks imitated from Cassia V1.0 and V2.0 datasets, which include a support and query set in each task. To counter severe class imbalance, especially the underrepresentation of tampered samples, two key enhancements are integrated: focal loss, which prioritizes hard-to-classify examples, and minority class oversampling, which synthetically boosts the tampered image population. A lightweight Convolutional Neural Network (CNN) is used as the backbone for both meta-learning and baseline evaluations. Extensive experimentation demonstrates that while the meta-model achieves high accuracy on authentic images, it struggles with tampered class recall due to underlying distribution skew. The baseline CNN, though lacking meta-adaptability, achieves modest recall on the tampered class. Quantitative results are presented using accuracy, precision, etc., and PR curve analysis. Visual and numerical outputs confirm that integrating meta-learning with class-aware strategies offers promising groundwork for robust, few-shot forgery detection. This study contributes a reproducible training pipeline and benchmark comparison that can be extended to more complex, real-world forgery scenarios in future research.

Keywords:

Image forgery detection, Meta-Learning, MAML, CASIA dataset, Focal loss, Class imbalance, Few-shot learning, Tampered images, CNN, Deep Learning.

References:

[1] Tahira Nazir et al., “Copy Move Forgery Detection and Segmentation Using Improved Mask Region-Based Convolution Network (RCNN),” Applied Soft Computing, vol. 131, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[2] S.B.G. Tilak Babu, and Ch Srinivasa Rao, “An Optimized Technique for Copy - Move Forgery Localization using Statistical Features,” ICT Express, vol. 8, no. 2, pp. 244-249, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[3] Yaqi Liu et al., “Two-Stage Copy-Move Forgery Detection With Self Deep Matching and Proposal SuperGlue,” IEEE Transactions on Image Processing, vol. 31, pp. 541-555, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[4] N. Krishnaraj et al., “Retracted: Design of Automated Deep Learning‐Based Fusion Model for Copy‐Move Image Forgery Detection,” Computational Intelligence and Neuroscience, vol. 2022, no. 1, pp. 1-14, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[5] Esteban Alejandro Armas Vega et al., “Copy-Move Forgery Detection Technique based on Discrete Cosine Transform Blocks Features,” Neural Computing and Applications, vol. 33, pp. 4713-4727, 2021.
[CrossRef] [Google Scholar] [Publisher Link]
[6] Haipeng Chen, Xiwen Yang, and Yingda Lyu, “Copy-Move Forgery Detection based on Keypoint Clustering and Similar Neighborhood Search Algorithm,” IEEE Access, vol. 8, pp. 36863-36875, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[7] Richa Singh et al., “Copy-Move Forgery Detection using SIFT and DWT Detection Techniques,” 2022 3rd International Conference on Intelligent Engineering and Management (ICIEM), London, United Kingdom, pp. 338-343, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[8] Patrick Niyishaka, and Chakravarthy Bhagvati, “Copy-Move Forgery Detection using Image Blobs and BRISK Feature,” Multimedia Tools and Applications, vol. 79, pp. 26045-26059, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[9] Guiwei Fu, Yujin Zhang, and Yongqi Wang, “Image Copy-Move Forgery Detection Based on Fused Features and Density Clustering,” Applied Sciences, vol. 13, no. 13, pp. 1-20, 2023.
[CrossRef] [Google Scholar] [Publisher Link]
[10] Yıldız Aydın, “Automated Identification of Copy-Move Forgery Using Hessian and Patch Feature Extraction Techniques,” Journal of Forensic Sciences, vol. 69, no. 1, pp. 131-138, 2024.
[CrossRef] [Google Scholar] [Publisher Link]
[11] S. Shashikala, and G.K. Ravi Kumar, “Optimization of Copy-Move Forgery Detection with Region Selection based on Domain Specific Characteristics,” International Journal of Intelligent Systems and Applications in Engineering, vol. 11, no. 4, pp. 693-702, 2023.
[Publisher Link]
[12] D. Shine Rajesh et al., “Image Forensic using ORB Algorithm for Digital Image Copy Move Forgery Detection,” International Journal of Computing and Artificial Intelligence, vol. 5, no. 2, pp. 95-98, 2024.
[CrossRef] [Publisher Link]
[13] Divya Prathana Timothy, and Ajit Kumar Santra, “Deep Learning-Based Copy-Move and Spliced Image Forgery Detection,” International Journal on Recent and Innovation Trends in Computing and Communication, vol. 11, no. 10, pp. 854-861, 2023.
[CrossRef] [Publisher Link]
[14] Yingjie He et al., “Image Copy-Move Forgery Detection via Deep Cross-Scale Patchmatch,” 2023 IEEE International Conference on Multimedia and Expo (ICME), Brisbane, Australia, pp. 2327-2332, 2023.
[CrossRef] [Google Scholar] [Publisher Link]
[15] Yuanman Li et al., “Image Copy-Move Forgery Detection via Deep PatchMatch and Pairwise Ranking Learning,” IEEE Transactions on Image Processing, vol. 34, pp. 425-440, 2025.
[CrossRef] [Google Scholar] [Publisher Link]
[16] Jingyu Wang et al., “Object-Level Copy-Move Forgery Image Detection based on Inconsistency Mining,” Companion Proceedings of the ACM Web Conference 2024, pp. 943-946, 2024.
[CrossRef] [Google Scholar] [Publisher Link]
[17] Li Jiang, and Zhaowei Lu, “An Effective Image Copy-Move Forgery Detection using Entropy Information,” 2024 9th International Conference on Image, Vision and Computing (ICIVC), Suzhou, China, pp. 344-349, 2024.
[CrossRef] [Google Scholar] [Publisher Link]
[18] Shizhen Chang, “Can Deep Network Balance Copy-Move Forgery Detection and Distinguishment?,” arXiv Preprint, pp. 1-7, 2023.
[CrossRef] [Google Scholar] [Publisher Link]
[19] Yaqi Liu et al., “CMFDFormer: Transformer-based Copy-Move Forgery Detection with Continual Learning,” arXiv Preprint, pp. 1-12, 2023.
[CrossRef] [Google Scholar] [Publisher Link]
[20] V. NavyaSree et al., “Predicting the Risk Factor of Kidney Disease using Meta Classifiers,” 2022 IEEE 2nd Mysore Sub Section International Conference (MysuruCon), Mysuru, India, pp. 1-6, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[21] Srinivas Naik et al., “Efficient Diabetic Retinopathy Detection using Convolutional Neural Network and Data Augmentation,” Soft Computing, vol. 28, 2024.
[CrossRef] [Google Scholar] [Publisher Link]