A Novel Approach for Detection and Classification of Fake Image

The rise of social media and the easy access to digital editing tools have greatly increased the number of manipulated and fake images on the internet. Since visual content is essential for communication, forming opinions, and sharing information, the authenticity of images has become important for public trust, cybersecurity, and social stability. Fake images can fuel misinformation campaigns, threaten national security, sway public opinion, and skew legal or political processes. The extent and complexity of these manipulations make it harder for people or traditional systems to detect them reliably. This survey paper gives an overview of the machine learning and deep learning methods used to detect and classify fake images, particularly focusing on Convolutional Neural Networks (CNNs). Unlike older methods that require feature engineering, CNNs automatically learn visual features and identify subtle inconsistencies caused by image tampering. The literature reviewed in this survey shows a clear shift from traditional methods, such as metadata analysis, noise estimation, and handcrafted features, to more sophisticated deep learning models that can detect complex manipulations. Despite significant advances, the field still has challenges. The shortage of large and varied real-world datasets, rapid progress in generative AI (like GANs), and the demands of deep learning models present major hurdles. Future research should improve the adaptability, strength, and clarity of detection models to effectively address increasingly sophisticated forgeries across various online spaces. The comparative evaluation shows that ResNet-50 achieved the highest overall classification performance, obtaining an accuracy of 97.5% and an F1-score of 97.1%. This indicates that ResNet-50 was the most effective architecture for capturing complex forgery-related visual patterns in manipulated images. DenseNet121 also performed strongly, achieving 96.9% accuracy, which demonstrates its capability to preserve and reuse fine-grained features useful for detecting subtle image manipulations. EfficientNet-B0, while slightly lower in classification accuracy (96.8%), showed the fastest inference time (12 ms) among all tested models. This makes it a suitable option for real-time or resource-constrained deployment scenarios where computational efficiency is important. Overall, the results suggest that ResNet-50 is best suited for maximum detection performance, while DenseNet121 and EfficientNet-B0 provide strong alternative choices depending on deployment requirements.

[1]                  D. P. Timothy and A. K. Santra, "Detecting Digital Image Forgeries with Copy-Move and Splicing Image Analysis using Deep Learning Techniques," Int. J. Adv. Comput. Sci. Appl. (IJACSA), vol. 15, no. 5, pp. 1–10, 2024. [2]                  M. Abdelmaksoud, B. Youssef, K. Wassif, and R. A. El-Khoribi, "Hybrid Framework for Image Forgery Detection and Robustness Against Adversarial Attacks Using Vision Transformer and SVM," Sci. Rep., vol. 15, Art. no. 40371, Nov. 2025. [3]                  A. Alabrah, "SFTA-Net: A Self-Supervised Approach to Detect Copy-Move and Splicing Forgery to Leverage Triplet Loss, Auxiliary Loss, and Spatial Attention," PeerJ Comput. Sci., vol. 11, e2803, Apr. 2025. [4]                  S. Radhakrishnan et al., "Exploring Machine Learning Approaches for Efficient Image Forgery Detection," J. Forensic Sci., vol. 70, no. 3, pp. 1–14, May 2025. [5]                  N. B. Naidu, T. Kavyasree, T. R. Teja, P. S. Sarayu, and S. Sai, "Image Forgery Detection Using ResNet50," Int. J. Res. Appl. Sci. Eng. Technol. (IJRASET), vol. 12, no. 3, pp. 1–7, 2024. [6]                  H. R. Suresh et al., "Deep Learning-Based Image Forgery Detection System," Int. J. Electron. Secur. Digit. Forensics, vol. 16, no. 2, pp. 160–172, 2024. [7]                  F. Z. Mehrjardi, A. M. Latif, M. S. Zarchi, and R. Sheikhpour, "A Survey on Deep Learning-Based Image Forgery Detection," Pattern Recognit., vol. 144, Art. no. 109778, 2023, doi: 10.1016/j.patcog.2023. [8]                  S. Agarwal, S. Walia, and K.-H. Jung, "A Cohesive Forgery Detection for Splicing and Copy-Paste in Digital Images," Multimed. Tools Appl., vol. 84, pp. 1–17, 2024, doi: 10.1007/s11042-024-18154-7.. [9]                  A. Akram et al., "Weber Law Based Approach for Multi-Class Image Forgery Detection," Comput. Mater. Contin. (CMC), vol. 78, no. 1, pp. 145–166, 2024, doi: 10.32604/cmc.2023.041074. [10]                (CNN + ELA + Dual-Stream on CASIA, March 2025) "Deep Learning-Based Image Forgery Detection Using CNN and Dual-Stream UNet," IJARCCE, vol. 14, no. 3, Mar. 2025.