Markov processes for enhanced deepfake generation and detection

Michael A. Kouritzin; Ian Zhang; Jyoti Bhadana; Seoyeon Park; Michael A. Kouritzin; Ian Zhang; Jyoti Bhadana; Seoyeon Park

doi:10.3934/math.2026483

AIMS Mathematics

2026, Volume 11, Issue 4: 11731-11759. doi: 10.3934/math.2026483

Previous Article Next Article

Research article Special Issues

Markov processes for enhanced deepfake generation and detection

1.
Department of Mathematical and Statistical Sciences, University of Alberta, Edmonton, AB, Canada
2.
Department of Statistical Sciences, University of Toronto, Toronto, ON, Canada
3.
Department of Mathematics, University of Texas at Arlington, Arlington, TX, USA

Received: 11 December 2025 Revised: 12 March 2026 Accepted: 03 April 2026 Published: 28 April 2026
MSC : 62M05; 60J22; 65C40

We investigate both new and existing methods for generating, and especially detecting, deepfakes through the simple but informative task of authenticating binary coin flip data. The main contribution is the introduction of a Markov observation model (MOM) as an alternative probabilistic framework for both deepfake generation and discrimination. Its performance is compared against several existing approaches, such as generative adversarial networks (GANs), support vector machines (SVMs), branching particle filtering (BPF), and human alternatives. Since SVMs are discriminative methods and do not have generative abilities, they are only evaluated for the detection task, while the remaining approaches are assessed on both generation and discrimination. Across the experiments, human participants are shown to perform the worst, which demonstrates the difficulty of reliably identifying deepfaked sequences by a human eye. Among the computational methods, GANs perform better than humans, but are outperformed by SVMs, which in turn are surpassed by BPF. The strongest overall performance comes from the proposed MOM approach, which achieves the best results for deepfake detection out of all methods. A similar result is observed for the generation task, with MOM again showing the strongest performance, followed by BPF, GAN, and humans. These results showcase the generative and discrimination abilities of the proposed method.
- deepfakes,
- generative adversarial networks,
- Markov chain,
- detection,
- likelihood,
- model selection,
- simulation
Citation: Michael A. Kouritzin, Ian Zhang, Jyoti Bhadana, Seoyeon Park. Markov processes for enhanced deepfake generation and detection[J]. AIMS Mathematics, 2026, 11(4): 11731-11759. doi: 10.3934/math.2026483

Related Papers:

Abstract

We investigate both new and existing methods for generating, and especially detecting, deepfakes through the simple but informative task of authenticating binary coin flip data. The main contribution is the introduction of a Markov observation model (MOM) as an alternative probabilistic framework for both deepfake generation and discrimination. Its performance is compared against several existing approaches, such as generative adversarial networks (GANs), support vector machines (SVMs), branching particle filtering (BPF), and human alternatives. Since SVMs are discriminative methods and do not have generative abilities, they are only evaluated for the detection task, while the remaining approaches are assessed on both generation and discrimination. Across the experiments, human participants are shown to perform the worst, which demonstrates the difficulty of reliably identifying deepfaked sequences by a human eye. Among the computational methods, GANs perform better than humans, but are outperformed by SVMs, which in turn are surpassed by BPF. The strongest overall performance comes from the proposed MOM approach, which achieves the best results for deepfake detection out of all methods. A similar result is observed for the generation task, with MOM again showing the strongest performance, followed by BPF, GAN, and humans. These results showcase the generative and discrimination abilities of the proposed method.

References

[1]	D. J. Ballantyne, H. Y. Chan, M. A. Kouritzin, Novel branching particle method for tracking, Signal and Data Processing of Small Targets 2000, 4048 (2000), 277–287. https://doi.org/10.1117/12.391984 doi: 10.1117/12.391984
[2]	T. Baltrusaitis, C. Ahuja, L. P. Morency, Multimodal machine learning: a survey and taxonomy, IEEE Trans. Pattern Anal. Mach. Intell., 41 (2019), 423–443. https://doi.org/10.1109/tpami.2018.2798607 doi: 10.1109/tpami.2018.2798607
[3]	C. Batanero, E. Sanchez, What is the nature of high school students' conceptions and misconceptions about probability? In: G. A. Jones, Exploring probability in school: challenges for teaching and learning, Mathematics Education Library, Springer US, Boston, MA, 2005,241–266. https://doi.org/10.1007/0-387-24530-8_11
[4]	O. Cappe, S. J. Godsill, E. Moulines, An overview of existing methods and recent advances in sequential Monte Carlo, Proceedings of the IEEE, 95 (2007), 899–924. https://doi.org/10.1109/JPROC.2007.893250 doi: 10.1109/JPROC.2007.893250
[5]	M. Costa, L. De Angelis, Model selection in hidden markov models: a simulation study, Quaderni di Dipartimento 7, Department of Statistics, University of Bologna, 2010.
[6]	D. Crisan, T. Lyons, Nonlinear filtering and measure-valued processes, Probab. Theory Relat. Fields, 109 (1997), 217–244. https://doi.org/10.1007/s004400050131 doi: 10.1007/s004400050131
[7]	P. Del Moral, L. Miclo. Branching and interacting particle systems approximations of feynman-kac formulae with applications to non-linear filtering, In: J. Azéma, M. Ledoux, M. Émery, M. Yor, Séminaire de probabilités XXXIV, Berlin, Heidelberg: Springer, 1729 (2000), 1–145. https://doi.org/10.1007/bfb0103798
[8]	R. Douc, O. Cappe, Comparison of resampling schemes for particle filtering, Proceedings of the 4th International Symposium on Image and Signal Processing and Analysis, Zagreb, Croatia, 2005, 64–69. https://doi.org/10.1109/ispa.2005.195385
[9]	A. Doucet, N. Freitas, N. Gordon, Sequential Monte Carlo methods in practice, New York, NY: Springer, 2001. https://doi.org/10.1007/978-1-4757-3437-9
[10]	I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, et al., Generative adversarial networks, Commun. ACM, 63 (2020), 139–144. https://doi.org/10.1145/3422622 doi: 10.1145/3422622
[11]	I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, et al., Generative adversarial nets, In: Advances in Neural Information Processing Systems, NIPS 2014, Volume 27, Curran Associates, Inc., 2014.
[12]	J. D. Hol, T. B. Schon, F. Gustafsson, On resampling algorithms for particle filters, 2006 IEEE Nonlinear Statistical Signal Processing Workshop, 2006, 79–82. https://doi.org/10.1109/nsspw.2006.4378824
[13]	K. Van Koevering, J. Kleinberg, How random is random? Evaluating the randomness and humaness of LLMs' coin flips, arXiv, 2024. https://doi.org/10.48550/arXiv.2406.00092
[14]	M. A. Kouritzin, Convergence rates for residual branching particle filters, J. Math. Anal. Appl., 449 (2017), 1053–1093. https://doi.org/10.1016/j.jmaa.2016.12.046 doi: 10.1016/j.jmaa.2016.12.046
[15]	M. A. Kouritzin, Residual and stratified branching particle filters, Comput. Stat. Data Anal., 111 (2017), 145–165. https://doi.org/10.1016/j.csda.2017.02.003 doi: 10.1016/j.csda.2017.02.003
[16]	M. A. Kouritzin, Markov observation models and deepfakes, Mathematics, 13 (2025), 2128. https://doi.org/10.3390/math13132128 doi: 10.3390/math13132128
[17]	M. A. Kouritzin, F. Newton, S. Orsten, D. C. Wilson, On detecting fake coin flip sequences, Inst. Math. Stat. (IMS) Collect., 4 (2008), 107–122. https://doi.org/10.1214/074921708000000336 doi: 10.1214/074921708000000336
[18]	M. A. Kouritzin, F. Newton, B. Wu, A flexible, real-time algorithm for simulating correlated random fields and its properties, J. Math. Stat., 13 (2017), 197–208. https://doi.org/10.3844/jmssp.2017.197.208 doi: 10.3844/jmssp.2017.197.208
[19]	L. Li, J. Bao, T. Zhang, H. Yang, D. Chen, F. Wen, et al., Face X-ray for more general face forgery detection, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 2020, 5000–5009. https://doi.org/10.1109/cvpr42600.2020.00505
[20]	Y. Li, X. Yang, P. Sun, H. Qi, S. Lyu, Celeb-DF: A large-scale challenging dataset for deepFake forensics, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 2020, 3204–3213. https://doi.org/10.1109/cvpr42600.2020.00327
[21]	L. M. Murray, A. Lee, P. E. Jacob, Parallel resampling in the particle filter, J. Comput. Graph. Stat., 25 (2016), 789–805. https://doi.org/10.1080/10618600.2015.1062015 doi: 10.1080/10618600.2015.1062015
[22]	A. Odena, C. Olah, J. Shlens, Conditional image synthesis with auxiliary classifier GANs, Proceedings of the 34th International Conference on Machine Learning, 2017, 2642–2651.
[23]	W. Pieczynski, Pairwise markov chains, IEEE Trans. Pattern Anal. Mach. Intell., 25 (2003), 634–639. https://doi.org/10.1109/tpami.2003.1195998 doi: 10.1109/tpami.2003.1195998
[24]	L. R. Rabiner, A tutorial on hidden Markov models and selected applications in speech recognition, Proc. IEEE, 77 (1989), 257–286. https://doi.org/10.1109/5.18626 doi: 10.1109/5.18626
[25]	A. Renelle, S. Budgett, E. Chernoff, Making heads and tails of generation loss: a timeless tale of folk randomness, Proceedings of the Eleventh International Conference on Teaching Statistics, 2023.
[26]	P. Révész, Strong theorems on coin tossing, Proceedings of the International Congress of Mathematicians (Helsinki, 1978), 2 (1980), 749–754.
[27]	A. Rossler, D. Cozzolino, L. Verdoliva, C. Riess, J. Thies, M. Niessner, FaceForensics++: Learning to detect manipulated facial images, 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea (South), 2019, 1–11. https://doi.org/10.1109/iccv.2019.00009
[28]	M. F. Schilling, The longest run of heads, Coll. Math. J., 21 (1990), 196–207. https://doi.org/10.1080/07468342.1990.11973306 doi: 10.1080/07468342.1990.11973306
[29]	T. Taulli, The impact on major industries, In: T. Taulli, Generative AI: How ChatGPT and other AI tools will revolutionize business, Berkeley, CA: Apress, 2023,175–188. https://doi.org/10.1007/978-1-4842-9367-6_8
[30]	T. Wang, X. Liao, K. P. Chow, X. Lin, Y. Wang, Deepfake detection: A comprehensive survey from the reliability perspective, ACM Comput. Surv., 57 (2024), 1–35. https://doi.org/10.1145/3699710 doi: 10.1145/3699710
[31]	P. A. Warren, U. Gostoli, G. D. Farmer, W. El-Deredy, U. Hahn, A re-examination of "bias'' in human randomness perception, J. Exp. Psychol.: Human Percept. Perform., 44 (2018), 663–680. https://doi.org/10.1037/xhp0000462 doi: 10.1037/xhp0000462
[32]	M. Westerlund, The emergence of deepfake technology: a review, Technol. Innov. Manag. Rev., 9 (2019), 39–52. https://doi.org/10.22215/timreview/1282 doi: 10.22215/timreview/1282
[33]	J. Xiong, An introduction to stochastic filtering theory, Oxford: Oxford University Press, 2023. https://doi.org/10.1093/oso/9780199219704.001.0001
[34]	K. Yang, W. Y. Lin, M. Barman, F. Condessa, Z. Kolter, Defending multimodal fusion models against single-source adversaries, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, TN, USA, 2021, 3339–3348, https://doi.org/10.1109/cvpr46437.2021.00335
[35]	X. Zhu, Y. Wang, E. Cambria, I. Rida, J. S. López, L. Cui, et al., RMER-DT: Robust multimodal emotion recognition in conversational contexts based on diffusion and transformers, Inf. Fusion, 123 (2025), 103268. https://doi.org/10.1016/j.inffus.2025.103268 doi: 10.1016/j.inffus.2025.103268

Reader Comments

Your name:*

Email:*
© 2026 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)