International Journal of Innovative Research in Computer and Communication Engineering
ISSN Approved Journal | Impact factor: 8.771 | ESTD: 2013 | Follows UGC CARE Journal Norms and Guidelines
| Monthly, Peer-Reviewed, Refereed, Scholarly, Multidisciplinary and Open Access Journal | High Impact Factor 8.771 (Calculated by Google Scholar and Semantic Scholar | AI-Powered Research Tool | Indexing in all Major Database & Metadata, Citation Generator | Digital Object Identifier (DOI) |
| TITLE | AI Image Forensics: A Tool for Detecting AI Generated Images Using Metadata Comparison |
|---|---|
| ABSTRACT | The thing with these systems like Stable Diffusion, Midjourney and DALL-E is that they can make fake pictures that look really real. I mean Stable Diffusion, Midjourney and DALL-E are so good that you cannot tell if a picture is fake or a real photograph. This is a problem for people who try to figure out if a picture is real or not and, for courts that need to know if a picture can be used as evidence. Stable Diffusion, Midjourney and DALL-E are making it harder to trust pictures. Existing detection methods use deep learning-based pixel-level analysis and rarely provide forensic explainability. In the current study, we propose an efficient web application that allows for identifying AI-generated images through comparing EXIF metadata attributes between a legitimate original picture and an AI-generated one. In particular, the presented system employs ten EXIF metadata attributes including Camera Make, Camera Model, Lens Model, GPS Info, Software Tag, ISO Speed, F Number, Focal Length, Exposure Time, and Date Time. Using the Pillow package, we perform a side-by-side color comparison of metadata, which becomes the basis for our primary forensic evidence. Additional comparisons include visual resemblance, luminosity, contrast ratio, number of edges, and DCT perceptual hashing, resulting in calculating a confidence score. A color comparison heatmap generated with OpenCV shows areas of maximum difference. Risk assessment can be made at Low, Medium, and High level, depending on the confidence score value. Results are stored in SQLite database and are provided in the form of a detailed forensic report created with Report Lab. Experiments carried out on a dataset of 1,000 images generated by seven AI-image generator applications lead to 89.9% accuracy and 0.942 AUC-ROC without requiring GPU. Our results show that 99.6% of AI-generated images have no GPS info and 98.6% – empty or minimal EXIF structures. |
| AUTHOR | SANTHAKUMAR M G, SANKARA NARAYANAN S T PG Student, Dept. of C.F.I.S, Dr. M.G.R. Educational and Research Institute, Chennai, India Dept. of Cyber Security, Dr. MGR Educational and Research Institute, Chennai, India |
| VOLUME | 185 |
| DOI | DOI: 10.15680/IJIRCCE.2026.1406006 |
| pdf/6_AI Image Forensics A Tool for Detecting AI Generated Images Using Metadata Comparison.pdf | |
| KEYWORDS | |
| References | [1] R. F. Astillero, ‘Forensic Analysis of Image Metadata to Distinguish AI-Generated Images,’ Colegio de Montalban, Philippines, 2025. [Online]. Available: https://www.researchgate.net/publication/394477131 [2] C. J. Walker, ‘Digital Forensics and AI: Artifact Analysis and Using AI in the Forensics Domain,’ LSU Doctoral Dissertations, 6884, 2025. [3] C. Kim, ‘Distinguishing AI-Generated and Real Images Using Swin-Transformer,’ 2025 IEEE Integrated STEM Education Conference (ISEC), Princeton, NJ, USA, pp. 1–6, doi: 10.1109/ISEC64801.2025.11147328. [4] I. Goodfellow et al., ‘Generative Adversarial Nets,’ Advances in Neural Information Processing Systems, vol. 27, pp. 2672–2680, 2014. [5] N. Mansoor and A. I. Iliev, ‘Explainable AI for DeepFake Detection,’ Applied Sciences, vol. 15, no. 2, p. 725, 2025, doi: 10.3390/app15020725. [6] A. Ara, M. S. Alam and A. F. Mifa, ‘A Comparative Review of AI-Generated Image Detection Across Social Media Platforms,’ Global Mainstream Journal of Innovation, Engineering and Emerging Technology, vol. 3, no. 1, pp. 11–22, 2024. [7] J. Madake, J. Meshram, A. Mondhe and P. Mashalkar, ‘Image Tampering Detection Using Error Level Analysis and Metadata Analysis,’ 2023 4th INCET, Belgaum, India, pp. 1–7, doi: 10.1109/INCET57972.2023.10169948. [8] C. Buangam, M. Phaiphon, N. Jewmungme and P. Worrawichaipat, ‘Detection of AI-Generated vs. Real Human Images Using Anatomical Subregions,’ 2025 29th ICSEC, Chiang Mai, Thailand, pp. 36–42. [9] J. Lukas, J. Fridrich and M. Goljan, ‘Digital Camera Identification from Sensor Pattern Noise,’ IEEE Transactions on Information Forensics and Security, vol. 1, no. 2, pp. 205–214, 2006. [10] J. Frank et al., ‘Leveraging Frequency Analysis for Deep Fake Image Recognition,’ Proc. 37th ICML, vol. 119, pp. 3247–3258, 2020. [11] M. Tan and Q. V. Le, ‘EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,’ Proc. 36th ICML, vol. 97, pp. 6105–6114, 2019. [12] W. Yang, R. Fu, M. B. Amin and B. Kang, ‘The Impact of Modern AI in Metadata Management,’ Human-Centric Intelligent Systems, vol. 5, pp. 323–350, 2025, doi: 10.1007/s44230-025-00106-5. |
