Model ConvNext Berbasis Fusi Fitur Spektral untuk Deteksi Citra Sintetis yang Robust

Authors

  • Reyvan Revolusioner Ar Universitas Halu Oleo
  • Adha Mashur Sajiah Universitas Halu Oleo
  • Sutardi Sutardi Universitas Halu Oleo

DOI:

https://doi.org/10.69693/ijmst.v5i1.14703

Keywords:

ConvNeXt, Deteksi Citra Sintetis, Fusi Fitur Spektral, Robustness, Adaptasi Domain

Abstract

Model generatif semakin mampu menghasilkan citra realistis sehingga deteksi citra sintetis menjadi sulit setelah citra mengalami kompresi, pengaburan, derau, atau perubahan ukuran. Penelitian ini mengembangkan detektor ConvNeXt-Tiny yang menggabungkan fitur spasial dengan lima representasi forensik: magnitudo Fourier, discrete cosine transform (DCT), wavelet Haar, fase Fourier, dan residual spatial rich model (SRM). Keenam cabang diproyeksikan menjadi embedding 256 dimensi dan diintegrasikan melalui gated residual fusion untuk menghasilkan kepala klasifikasi spatial, spectral, dan fusion. Evaluasi eksperimen mencakup klasifikasi bersih, 21 kondisi pascapemrosesan, adaptasi domain baru, regresi domain lama, dan pengujian white box. Pada benchmark model dasar, fusion memperoleh akurasi 93,00% pada citra bersih dan mengungguli spatial dalam 18 dari 22 kondisi. Keunggulan terbesar sebesar 1,69 poin persentase terjadi pada resize 0,05, sedangkan pada noise Gaussian sigma 0,10 fusion lebih rendah 1,50 poin persentase. Fine-tuning memilih epoch ketujuh dengan skor validasi gabungan 0,7991. Pada 263 citra uji domain baru yang dinormalisasi, akurasi fusion meningkat dari 53,23% menjadi 84,79% dan macro-F1 dari 0,4416 menjadi 0,8473. Uji McNemar exact menunjukkan perubahan sebelum dan sesudah fine-tuning yang signifikan (p < 0,001). Namun, akurasi pada 2.000 citra domain lama menurun dari 84,10% menjadi 80,55%. Sebelas kasus white box lulus, termasuk empat jalur dasar fungsi fusi. Hasil menunjukkan bahwa manfaat fusi spektral bergantung pada kondisi gangguan dan adaptasi domain menimbulkan kompromi antardomain. Ablation terkontrol dan perbandingan berpasangan antarkepala masih diperlukan untuk memastikan kontribusi cabang serta keunggulan statistik fusion terhadap spatial.

References

1. Alam, I., Muneer, M. S., & Woo, S. S. (2024). UGAD: Universal generative AI detector utilizing frequency fingerprints. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management (pp. 4332–4340). https://doi.org/10.1145/3627673.3680085

2. Amarasinghe, H. M. S., & Arachchi, S. P. K. (2025). Deepfake detection using a hybrid deep learning approach with Swin Transformers and ConvNeXt. In Proceedings of the International Research Conference on Smart Computing and Systems Engineering (SCSE 2025). https://doi.org/10.1109/SCSE65633.2025.11030955

3. Bergmann, S., Moussa, D., Brand, F., Kaup, A., & Riess, C. (2024). Forensic analysis of AI-compression traces in spatial and frequency domain. Pattern Recognition Letters, 180, 41–47. https://doi.org/10.1016/j.patrec.2024.02.015

4. Bhattacharjee, A., Islam, K., Anan, K., Intesher, A., Fuad, A. A., Saha, U., & Imtiaz, H. (2026). CAE-Net: Generalized deepfake image detection using convolution and attention mechanisms with spatial and frequency domain features. Journal of Visual Communication and Image Representation, 115, 104679. https://doi.org/10.1016/j.jvcir.2025.104679

5. Corvi, R., Cozzolino, D., Zingarini, G., Poggi, G., Nagano, K., & Verdoliva, L. (2023). On the detection of synthetic images generated by diffusion models. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 1–5). https://doi.org/10.1109/ICASSP49357.2023.10095167

6. Hardiyanto, A., & Akbar, M. (2025). Klasifikasi citra sintetis hasil model difusi menggunakan Gray Level Co-Occurrence Matrix (GLCM) dan Convolutional Neural Network (CNN). JIKO (Jurnal Informatika dan Komputer), 9(3), 572–582. https://doi.org/10.26798/jiko.v9i3.2033

7. Ilyas, H., Javed, A., & Malik, K. M. (2024). ConvNext-PNet: An interpretable and explainable deep-learning model for deepfakes detection. In 2024 IEEE International Joint Conference on Biometrics (IJCB). https://doi.org/10.1109/IJCB62174.2024.10744484

8. Liu, C., Zhu, T., Zhou, W., & Zhao, W. (2026). Frequency bias matters: Diving into robust and generalized deep image forgery detection. IEEE Transactions on Dependable and Secure Computing, 23(2), 3636–3651. https://doi.org/10.1109/TDSC.2025.3639022

9. Liu, Z., Mao, H., Wu, C.-Y., Feichtenhofer, C., Darrell, T., & Xie, S. (2022). A ConvNet for the 2020s. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 11976–11986). https://doi.org/10.1109/CVPR52688.2022.01167

10. Nourji, S., Subramanian, T., Pakala, S., & Yang, E. (2025). Evaluating convolutional neural networks for synthetic image detection in the frequency domain. In 2025 IEEE International Conference on AI and Data Analytics (ICAD). https://doi.org/10.1109/ICAD65464.2025.11114037

11. Raharjo, T., Adristi, F. I., Romadhona, F. Y., Syahputra, R. R., Halim, M. Y., Rachman, M. A., Marjianto, R. N., Santicho, D., Kusuma, P., & Ramadhani, E. (2025). Analisis forensik deepfake berbasis Convolutional Neural Network (CNN) untuk deteksi inkonsistensi tekstur dan pola pada citra wajah. JATI (Jurnal Mahasiswa Teknik Informatika), 9(2), 2731–2738. https://doi.org/10.36040/jati.v9i2.13058

12. Ricker, J., Damm, S., Holz, T., & Fischer, A. (2024). Towards the detection of diffusion model deepfakes. In Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISAPP) (Vol. 4, pp. 446–457). https://doi.org/10.5220/0012422000003660

13. Taufiqurrahman, R. F., Anggraeny, F. T., & Sihananto, A. N. (2025). Penerapan arsitektur CNN-Dilated untuk deteksi manipulasi citra deepfake. ILKOMNIKA: Journal of Computer Science and Applied Informatics, 7(3), 495–506. https://doi.org/10.28926/ilkomnika.v7i3.821

14. Wißmann, A., Zeiler, S., Nickel, R. M., & Kolossa, D. (2024). Whodunit: Detection and attribution of synthetic images by leveraging model-specific fingerprints. In Proceedings of the ACM Workshop on Multimedia AI against Disinformation (MAD ’24) (pp. 65–72). https://doi.org/10.1145/3643491.3660280

15. Woo, S., Debnath, S., Hu, R., Chen, X., Liu, Z., Kweon, I. S., & Xie, S. (2023). ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 16133–16142). https://doi.org/10.1109/CVPR52729.2023.01548

16. Zhao, H., Wei, T., Zhou, W., Zhang, W., Chen, D., & Yu, N. (2021). Multi-attentional deepfake detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 2185–2194). https://doi.org/10.1109/CVPR46437.2021.00222

Downloads

Published

16-11-2026

How to Cite

Ar, R. R., Sajiah, A. M., & Sutardi, S. (2026). Model ConvNext Berbasis Fusi Fitur Spektral untuk Deteksi Citra Sintetis yang Robust . Indonesian Journal of Multidisciplinary on Social and Technology, 5(1), 1153–1162. https://doi.org/10.69693/ijmst.v5i1.14703