Mathematics and Computational Sciences

Generative adversarial network-based forensic facial reconstruction from 3D face meshes: a comparative study of DCGAN and StyleGAN

Document Type : Original Article

Authors

Thiagarajan Kalyani 1
Gurusamy Meena Devi 2
Murali Gayathri Lakshmi 3
Immanuel Prabaharan Soundararaja 4

1 Department of Mathematics St. Joseph’s Institute of Technology Chennai, Tamil Nadu, India

2 Department of Mathematics, St. Joseph's College of Engineering, Chennai, Tamil Nadu, India

3 Department of Mathematics, Saveetha Engineering college, Chennai, Tamil Nadu, India

4 Department of Electrical & Electronics Engineering, AMET Deemed to be University, Chennai, India.

https://doi.org/10.30511/mcs.2026.2077705.1621
Abstract
Forensic facial reconstruction is a critical investigative technique for identifying unknown individuals from skeletal remains, facial meshes, or degraded photographs. Existing approaches manual clay reconstruction, drag-and-drop composite software, and Shape-from-Shading algorithms are time consuming, cost-prohibitive, or limited to well-defined facial inputs. This paper proposes an automated, generative AI based pipeline that reconstructs realistic facial images from 3-D face mesh representations and 2-D facial sketches using Deep Convolutional Generative Adversarial Networks (DCGAN) and Style-based Generative Adversarial Networks (StyleGAN). A dedicated dataset of 5 000 paired mesh-to-face image samples was constructed using Google MediaPipe, covering diverse age groups, ethnicities, and genders with an 80/10/10 train-validationtest split. DCGAN achieved a testing accuracy of 94.2% at 300 epochs, while StyleGAN attained a Fr´echet Inception Distance (FID) of 18.3 at 200 epochs with SSIM = 0.78 and PSNR = 28.1 dB. Comparative evaluation shows that DCGAN excels at rapid broad-pattern reconstruction, while StyleGAN provides superior feature-level perceptual fidelity. Ethical risks including misidentification, dataset consent, demographic bias, and legal implications are also addressed.

Keywords

Facial Reconstruction
StyleGAN
DCGAN
Forensic AIF
Subjects
[1] Brock, A., Donahue, J., & Simonyan, K. (2019). Large scale GAN training for high fidelity natural image synthesis. In Proceedings of the International Conference on Learning Representations (ICLR), New Orleans, LA. 
[2] Chen, X., Duan, Y., Houthooft, R., Schulman, J., Sutskever, I., & Abbeel, P. (2016). InfoGAN: Interpretable representation learning by information maximizing GANs. In Proceedings of the Conference on Neural Information Processing Systems (NeurIPS), Barcelona, Spain, pp. 2172–2180.
[3] Chen, Y., Yin, W., & Tang, X. (2019). 3D-aided face synthesis for pose-invariant recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41(10), 2418–2431.
[4] Chen, L., & Wang, Q. (2019). Shape-from-shading techniques for accurate 3D facial reconstruction: A survey. IEEE Computer Graphics and Applications, 39(2), 52–61.  
[5] Chen, Y., & Wu, H. (2021). GANs for facial image synthesis in forensics: A survey. Forensic Science International, 318, 110550.
[6] Das, P., Roy, M., & Kumar, S. (2019). Limitations and improvements in drag-and-drop facial feature systems for forensic composite generation. Forensic Science International, 302, 109899. 
[7] Das, S., & Roy, M. (2019). Facial reconstruction from skull images using GANs: An empirical study. Pattern Recognition Letters, 72, 30–38.
[8] Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative adversarial nets. In Proceedings of the Conference on Neural Information Processing Systems (NeurIPS), Montreal, Canada, pp. 2672–2680. 
[9] Gupta, R., & Verma, S. (2021). Forensic facial reconstruction: Challenges and opportunities in the era of deep learning.  Journal of Forensic Sciences, 66(3), 785–797.
[10] Isola, P., Zhu, J.-Y., Zhou, T., & Efros, A. A. (2017). Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, pp. 1125–1134.
[11] Kapoor, R., & Gupta, S. (2020). Deep learning approaches for facial feature enhancement in forensic reconstruction. IEEE Transactions on Image Processing, 29, 5789–5798. 
[12] Karras, T., Aila, T., Laine, S., & Lehtinen, J. (2018). Progressive growing of GANs for improved quality, stability, and variation. In Proceedings of the International Conference on Learning Representations (ICLR), Vancouver, Canada. 
[13] Karras, T., Laine, S., Aittala, M., Hellsten, J., Lehtinen, J., & Aila, T. (2020). Analyzing and improving the image quality  of StyleGAN. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, pp. 8107–8116.
[14] Kim, H., & Lee, S. (2016). Facial reconstruction in virtual reality environments: An IEEE survey. IEEE Computer Graphics and Applications, 36(4), 40–53.
[15] Kumar, A., & Das, S. (2015). Forensic facial reconstruction: A machine learning perspective. Pattern Recognition Letters, 58, 20–28.
[16] Richardson, E., Alaluf, Y., Patashnik, O., Nitzan, Y., Azar, Y., Shapiro, S., & Cohen-Or, D. (2021). Encoding in style: A StyleGAN encoder for image-to-image translation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, TN, pp. 2287–2296.
[17] Rodriguez, M., et al. (2017). Real-time applications of shape-from-shading in facial reconstruction. IEEE Transactions on Visualization and Computer Graphics, 23(11), 2456–2465. 
[18] Smith, J., & Johnson, A. (2018). Advancements in forensic facial reconstruction: A comprehensive review. IEEE Transactions on Biomedical Engineering, 65(7), 1543–1552.  
[19] Zhang, H., Xu, T., Li, H., Zhang, S., Wang, X., Huang, X., & Metaxas, D. (2017). StackGAN: Text to photo-realistic image synthesis with stacked GANs. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), Venice, Italy, pp. 5907–5915. 
[20] Zhang, L., Wang, Y., & Li, Q. (2021). Deep generative models for facial feature synthesis: A comprehensive review. IEEE Transactions on Affective Computing, 12(1), 1–15.
Mathematics and Computational Sciences
Volume 7, Issue 3
Summer 2026
Pages 160-167

  • Receive Date 14 November 2025
  • Revise Date 20 June 2026
  • Accept Date 20 June 2026

Home

Submit Manuscript

Reviewers

Contact Us