3D FACE IMAGE RECOGNITION USING QUATERNION NEURAL NETWORKS
DOI:
https://doi.org/10.28925/2663-4023.2026.35.1398Keywords:
3D face recognition, head pose estimation, quaternion neural networks (QCNN), Hamilton product, hybrid architectures, MobileNetV2, Huber Loss, L2 regularization, Edge AIAbstract
This article addresses a pressing scientific and applied problem in computer vision: improving the accuracy of 3D face orientation (pose) estimation and reducing the parametric complexity of deep neural network models through the use of quaternion algebra. The study examines the limitations of real-valued convolutional neural networks (CNNs), which process coordinates and color channels independently; this approach loses intrinsic inter-component correlations and leads to degraded accuracy in the presence of significant head rotations and occlusions.
The use of custom quaternion convolutional layers (QuaternionConv2D)—based on the non-commutative Hamilton product—is proposed and mathematically justified; this enables the modeling of spatial-angular signal interdependencies and reduces the number of free weight parameters by a factor of four compared to classical convolutions. To ensure stable model training, a weighted adaptive Huber loss function in the radian domain was developed—placing greater weight on the highly variable yaw angle—and L2 weight regularization was applied.
A systematic comparative experimental analysis was conducted on three types of architectures: a purely quaternion network (Stable QCNN) and a hybrid model (MobileNetV2 + QCNN Hybrid). Based on the results obtained, it has been demonstrated that the hybrid architecture achieves the best recognition accuracy, reducing the validation MAE to 0.1535 rad (8.80°)—an improvement of 27.39% over the baseline MobileNetV2 (12.12°)—with a negligible increase in parameter count of only 2.48%. Meanwhile, the ultra-lightweight, purely quaternion-based Stable QCNN model exhibits extreme compactness (0.04M parameters, 60.5 times fewer than MobileNetV2) and a 40.3% shorter training time, making it a promising solution for resource-constrained Edge AI devices.
Downloads
References
Asperti, A., & Filippini, D. (2023). Deep learning for head pose estimation: A survey. SN Computer Science, 4(4), Article 349. https://doi.org/10.1007/s42979-023-01796-z
Singh, S., Kumar, S., & Tripathi, B. K. (2025). A comprehensive analysis of quaternion deep neural networks: Architectures, applications, challenges, and future scope. Archives of Computational Methods in Engineering, 32(4), 2607–2634. https://doi.org/10.1007/s11831-024-10216-1
Altamirano-Gomez, G., & Gershenson, C. (2024). Quaternion convolutional neural networks: Current advances and future directions. Advances in Applied Clifford Algebras, 34(4), Article 42. https://doi.org/10.1007/s00006-024-01350-x
Zhang, A., Tay, Y., Zhang, S., Chan, A., Luu, A. T., Hui, S. C., & Fu, J. (2021). Beyond fully-connected layers with quaternions: Parameterization of hypercomplex multiplications with 1/n parameters (arXiv:2102.08597). arXiv. https://doi.org/10.48550/ARXIV.2102.08597
El Alami, A., Mesbah, A., Berrahou, N., Lakhili, Z., Berrahou, A., & Qjidaa, H. (2023). Quaternion discrete orthogonal Hahn moments convolutional neural network for color image classification and face recognition. Multimedia Tools and Applications, 82(21), 32827–32853. https://doi.org/10.1007/s11042-023-14866-4
Li, M., Huang, B., & Tian, G. (2022). A comprehensive survey on 3D face recognition methods. Engineering Applications of Artificial Intelligence, 110, Article 104669. https://doi.org/10.1016/j.engappai.2022.104669
Algabri, R., Shin, H., Abdu, A., Bae, J.-H., & Lee, S. (2025). WQuatNet: Wide range quaternion-based head pose estimation. Journal of King Saud University - Computer and Information Sciences, 37(3), Article 24. https://doi.org/10.1007/s44443-025-00034-1
Hempel, T., Abdelrahman, A. A., & Al-Hamadi, A. (2024). Toward robust and unconstrained full range of rotation head pose estimation. IEEE Transactions on Image Processing, 33, 2377–2387. https://doi.org/10.1109/TIP.2024.3378180
Parcollet, T., Morchid, M., & Linarès, G. (2020). A survey of quaternion neural networks. Artificial Intelligence Review, 53(4), 2957–2982. https://doi.org/10.1007/s10462-019-09752-1
García-Retuerta, D., Casado-Vara, R., Martín-del Rey, Á., De La Prieta, F., Prieto, J., & Corchado, J. M. (2020). Quaternion neural networks: State-of-the-art and research challenges. In C. Analide, P. Novais, D. Camacho, & H. Yin (Eds.), Intelligent Data Engineering and Automated Learning – IDEAL 2020 (Vol. 12490, pp. 456–467). Springer International Publishing. https://doi.org/10.1007/978-3-030-62365-4_43
Bill, J., Champagne, L., Cox, B., & Bihl, T. (2021). Meta-heuristic optimization methods for quaternion-valued neural networks. Mathematics, 9(9), Article 938. https://doi.org/10.3390/math9090938
Shahadat, N., & Maida, A. S. (2023). Enhancing ResNet image classification performance by using parameterized hypercomplex multiplication (arXiv:2301.04623). arXiv. https://doi.org/10.48550/ARXIV.2301.04623
El Alami, A., Berrahou, N., Lakhili, Z., Mesbah, A., Berrahou, A., & Qjidaa, H. (2022). Efficient color face recognition based on quaternion discrete orthogonal moments neural networks. Multimedia Tools and Applications, 81(6), 7685–7710. https://doi.org/10.1007/s11042-021-11669-3
Wen, C., & Qiu, Y. (2022). Color occlusion face recognition method based on quaternion non-convex sparse constraint mechanism. Sensors, 22(14), Article 5284. https://doi.org/10.3390/s22145284
Zhou, Y., Jin, L., Liu, H., & Song, E. (2021). Color facial expression recognition by quaternion convolutional neural network with Gabor attention. IEEE Transactions on Cognitive and Developmental Systems, 13(4), 969–983. https://doi.org/10.1109/TCDS.2020.3041642
Liu, L., Chen, C. L. P., & Li, S. (2022). Hallucinating color face image by learning graph representation in quaternion space. IEEE Transactions on Cybernetics, 52(1), 265–277. https://doi.org/10.1109/TCYB.2020.2979320
Zhao, Y., Birdal, T., Lenssen, J. E., Menegatti, E., Guibas, L., & Tombari, F. (2020). Quaternion equivariant capsule networks for 3D point clouds. In A. Vedaldi, H. Bischof, T. Brox, & J.-M. Frahm (Eds.), Computer Vision – ECCV 2020 (Vol. 12346, pp. 1–19). Springer International Publishing. https://doi.org/10.1007/978-3-030-58452-8_1
You, X., & Zhao, X. (2025). A 3D face recognition algorithm directly applied to point clouds. Biomimetics, 10(2), Article 70. https://doi.org/10.3390/biomimetics10020070
Hsu, H.-W., Wu, T.-Y., Wan, S., Wong, W. H., & Lee, C.-Y. (2019). QuatNet: Quaternion-based head pose estimation with multiregression loss. IEEE Transactions on Multimedia, 21(4), 1035–1046. https://doi.org/10.1109/TMM.2018.2866770
Liu, W., Zhou, L., & Chen, J. (2021). Face recognition based on lightweight convolutional neural networks. Information, 12(5), Article 191. https://doi.org/10.3390/info12050191
Gaudet, C. J., & Maida, A. S. (2018). Deep quaternion networks. In 2018 International Joint Conference on Neural Networks (IJCNN) (pp. 1–8). IEEE. https://doi.org/10.1109/IJCNN.2018.8489651
Li, X., Zhang, D., Li, M., & Lee, D.-J. (2023). Accurate head pose estimation using image rectification and a lightweight convolutional neural network. IEEE Transactions on Multimedia, 25, 2239–2251. https://doi.org/10.1109/TMM.2022.3144893
Liu, H., Fang, S., Zhang, Z., Li, D., Lin, K., & Wang, J. (2022). MFDNet: Collaborative poses perception and matrix Fisher distribution for head pose estimation. IEEE Transactions on Multimedia, 24, 2449–2460. https://doi.org/10.1109/TMM.2021.3081873
Zhou, H., Zhang, X., Zhang, C., & Ma, Q. (2023). Quaternion convolutional neural networks for hyperspectral image classification. Engineering Applications of Artificial Intelligence, 123, Article 106234. https://doi.org/10.1016/j.engappai.2023.106234
Martínez-Díaz, Y., Luhbach, P., Benitez-Garcia, G., & Hernandez-Sosa, D. (2021). Benchmarking lightweight face architectures on specific face recognition scenarios. Artificial Intelligence Review, 54(8), 6201–6244. https://doi.org/10.1007/s10462-021-09974-2
Boutros, F., Damer, N., Fang, M., Kirchbuchner, F., & Kuijper, A. (2021). MixFaceNets: Extremely efficient face recognition networks. In 2021 IEEE International Joint Conference on Biometrics (IJCB) (pp. 1–8). IEEE. https://doi.org/10.1109/IJCB52358.2021.9484374
Grant, B., & Wang, P. (2025). Quaternion approximation networks for enhanced image classification and oriented object detection (arXiv:2509.05512). arXiv. https://doi.org/10.48550/ARXIV.2509.05512
Devadas, R. M., Hiremani, V., Preethi, S. T., S, R., & Gujjar, P. (2025). Hypercomplex neural networks: Exploring quaternion, octonion, and beyond in deep learning. MethodsX, 15, Article 103644. https://doi.org/10.1016/j.mex.2025.103644
Grassucci, E., Zhang, A., & Comminiello, D. (2024). PHNNs: Lightweight neural networks via parameterized hypercomplex convolutions. IEEE Transactions on Neural Networks and Learning Systems, 35(6), 8293–8305. https://doi.org/10.1109/TNNLS.2022.3226772
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Євген Чичкарьов, Іван Веремійчук
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
