Classification of Real and AI-Generated Face Images Using a Frozen MobileNetV2 Transfer Learning Model with a Web Demonstration

Annur Farida; Ratri Enggar Pawening; Wahab Sya'roni

Classification of Real and AI-Generated Face Images Using a Frozen MobileNetV2 Transfer Learning Model with a Web Demonstration

Authors

(1) * Annur Farida

(Universitas Nurul Jadid)
        Indonesia
(2)  Ratri Enggar Pawening   (Universitas Nurul Jadid)
        Indonesia
(3)  Wahab Sya'roni   (Universitas Nurul Jadid)
        Indonesia
(*) Corresponding Author

Abstract

Image generators based on diffusion models can now produce faces that people can barely distinguish from photographs, which creates risks of impersonation and misinformation. This study examines how far a frozen, ImageNet-pretrained MobileNetV2 backbone with a small trainable classification head can separate real faces from faces generated by artificial intelligence, and it wraps the trained model in a Flask web application as a demonstration. The dataset contains 2,200 face images (1,050 real and 1,150 synthetic, produced by SDXL, FLUX.1 Pro, FLUX.1 Dev, and Gemini), split 70:15:15 into training, validation, and test subsets. The model has 2.42 million parameters, of which 164,353 were trained. On a held-out test set of 331 images drawn from the same pool, it reached an accuracy of 95.47% (95% confidence interval 92.7% to 97.2%), with F1-scores of 0.956 for synthetic and 0.953 for real faces; the 15 errors comprised 10 synthetic faces labelled real and 5 real faces labelled synthetic. Because the real images come mostly from one photographic collection and most synthetic images from one generated collection, the study cannot separate the detection of image generation from the recognition of the collection. Cross-source, per-generator, and baseline comparisons were not carried out, and training was run once. The results are therefore preliminary, and the main contribution is a transparent in-distribution baseline together with a working prototype.

Keywords

Deepfake detection; Synthetic image detection; Lightweight deep learning; Image classification; Flask

Full Text: PDF

References

I. Goodfellow et al., “Generative adversarial nets,” in Proc. Adv. Neural Inf. Process. Syst. (NIPS), vol. 27, Montréal, QC, Canada, 2014, pp. 2672–2680.

R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), New Orleans, LA, USA, 2022, pp. 10684–10695.

D. Podell et al., “SDXL: Improving latent diffusion models for high-resolution image synthesis,” 2023, arXiv:2307.01952.

J. Mu, M. Adrezo, and A. N. Haikal, “Identifikasi wajah asli dan buatan deepfake menggunakan metode convolutional neural network,” Teknika, vol. 13, no. 1, pp. 45–50, 2024, doi: 10.34148/teknika.v13i1.705.

A. Firc, K. Malinka, and P. Hanáček, “Deepfakes as a threat to a speaker and facial recognition: An overview of tools and attack vectors,” Heliyon, vol. 9, no. 4, art. no. e15090, 2023, doi: 10.1016/j.heliyon.2023.e15090.

M. Zhu et al., “GenImage: A million-scale benchmark for detecting AI-generated image,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) Datasets Benchmarks Track, vol. 36, 2023.

H. S. Abdul kareem and M. S. M. Altaei, “Detection of deep fake in face images using deep learning,” Wasit J. Comput. Math. Sci., vol. 1, no. 4, pp. 60–71, 2022, doi: 10.31185/wjcm.92.

Y. Patel et al., “An improved dense CNN architecture for deepfake image detection,” IEEE Access, vol. 11, pp. 22081–22095, 2023, doi: 10.1109/ACCESS.2023.3251417.

M. A. Arshed, S. Mumtaz, M. Ibrahim, C. Dewi, M. Tanveer, and S. Ahmed, “Multiclass AI-generated deepfake face detection using patch-wise deep learning model,” Computers, vol. 13, no. 1, art. no. 31, 2024, doi: 10.3390/computers13010031.

A. H. Soudy et al., “Deepfake detection using convolutional vision transformers and convolutional neural networks,” Neural Comput. Appl., vol. 36, no. 31, pp. 19759–19775, 2024, doi: 10.1007/s00521-024-10181-7.

M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobileNetV2: Inverted residuals and linear bottlenecks,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Salt Lake City, UT, USA, 2018, pp. 4510–4520.

J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “ImageNet: A large-scale hierarchical image database,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Miami, FL, USA, 2009, pp. 248–255.

M. F. Naufal and S. F. Kusuma, “Pendeteksi citra masker wajah menggunakan CNN dan transfer learning,” J. Teknol. Inf. Ilmu Komput., vol. 8, no. 6, pp. 1293–1300, 2021, doi: 10.25126/jtiik.2021865201.

U. Ojha, Y. Li, and Y. J. Lee, “Towards universal fake image detectors that generalize across generative models,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2023, pp. 24480–24489.

R. Corvi, D. Cozzolino, G. Zingarini, G. Poggi, K. Nagano, and L. Verdoliva, “On the detection of synthetic images generated by diffusion models,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), 2023, doi: 10.1109/ICASSP49357.2023.10095167.

S.-Y. Wang, O. Wang, R. Zhang, A. Owens, and A. A. Efros, “CNN-generated images are surprisingly easy to spot... for now,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2020, pp. 8695–8704.

A. Rössler, D. Cozzolino, L. Verdoliva, C. Riess, J. Thies, and M. Nießner, “FaceForensics++: Learning to detect manipulated facial images,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2019, pp. 1–11, doi: 10.1109/ICCV.2019.00009.

shreyanshpatel1, “130k real vs fake face,” Kaggle dataset. [Online]. Available: https://www.kaggle.com/datasets/shreyanshpatel1/130k-real-vs-fake-face

T. Karras, S. Laine, and T. Aila, “A style-based generator architecture for generative adversarial networks,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Long Beach, CA, USA, 2019, pp. 4401–4410.

L. Alzubaidi et al., “Review of deep learning: Concepts, CNN architectures, challenges, applications, future directions,” J. Big Data, vol. 8, art. no. 53, 2021, doi: 10.1186/s40537-021-00444-8.

M. Lin, Q. Chen, and S. Yan, “Network in network,” 2013, arXiv:1312.4400.

S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” in Proc. 32nd Int. Conf. Mach. Learn. (ICML), Lille, France, 2015, pp. 448–456.

N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfitting,” J. Mach. Learn. Res., vol. 15, no. 56, pp. 1929–1958, 2014.

D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. 3rd Int. Conf. Learn. Represent. (ICLR), San Diego, CA, USA, 2015.

S. A. Hicks et al., “On evaluation metrics for medical applications of artificial intelligence,” Sci. Rep., vol. 12, art. no. 5979, 2022, doi: 10.1038/s41598-022-09954-8.

D. J. Hand, P. Christen, and N. Kirielle, “F: An interpretable transformation of the F-measure,” Mach. Learn.*, vol. 110, no. 3, pp. 451–456, 2021, doi: 10.1007/s10994-021-05964-1.

Refbacks

There are currently no refbacks.

Journal of Intelligent Computing and Applied Informatics (JICAI)