Facial classification Using Hybrid Quantum Machine Learning

Roshan Babu Bandlapalli; Srinivas V Katakam; Jitendra Chougala; Ravi Kumar Kappagantu; Jayasri Dontabhaktuni

cs.CVSep 25, 2026

Facial classification Using Hybrid Quantum Machine Learning

Authors: Roshan Babu Bandlapalli, Srinivas V Katakam, Jitendra Chougala, Ravi Kumar Kappagantu, Jayasri Dontabhaktuni

Organizations: Department of Computer Science and Engineering, Mahindra University, Hyderabad, India · Lloyds Technology Centre, Hyderabad, India · Department of Physics, Mahindra University, Hyderabad, India

Abstract

Hybrid quantum methods have received limited study for resource-constrained facial biometrics. We present a hybrid quantum-classical facial recognition pipeline designed to run on standard computing hardware. Images undergo gamma correction, contrast enhancement, and principal component analysis before their features are encoded into an eight-qubit variational quantum classifier. Classical image matching then performs recognition. In experiments with 50,000 images, comprising 25,000 faces from CelebA and 25,000 non-face images from CIFAR-10, the method outperformed the reported CPU-trained FaceNet baseline in accuracy and training efficiency. The pipeline was also evaluated on GPU and quantum hardware. In an attendance monitoring deployment at Mahindra University in collaboration with Lloyds Technology Centre, CPU inference took 0.2 to 0.5 seconds per person, and the system remained robust to the use of spectacles. These findings support the feasibility of deploying hybrid quantum methods for facial recognition on existing CPU hardware.

Figures & tables

Figure 1: Hybrid Classical–Quantum Facial Biometrics Architecture illustrating the two-stage pipeline. Stage 1 performs classical preprocessing such as Haar cascade detection, ROI extraction, gamma correction, CLAHE enhancement, and PCA dimensionality reduction. Stage 2 applies quantum processing via angle embedding, multi-layer strong entanglement, and measurement for face/non-face classification, followed by database matching for authentication.

Figure 2: Single layer of Variational Quantum Classifier (VQC) anstaz for facial detection implemented on 8-qubit register (one of three variational layers illustrated). The circuit comprises three principal components: (i) Angle embedding , (ii) Strongly entangling layers , followed by nearest-neighbor circular CNOT entanglement to promote feature correlation across the register; and (iii) Measurement stage , where Pauli- $Z$ expectation values are extracted from selected qubits for binary classification. The complete architecture contains 72 trainable parameters ( $3\times 8\times 3$ rotations) and operates within a $2^{8}=256$ -dimensional Hilbert space. The overall circuit depth remains approximately 25 gates per layer, ensuring compatibility with noisy intermediate-scale quantum (NISQ) hardware constraints.

Figure 3: Training and validation dynamics over 100 epochs showing unusual learning behavior. a: Loss curves demonstrate smooth convergence from 0.75 to approximately 0.30 without signs of overfitting—training and validation losses track closely throughout. b: Accuracy evolution reveals interesting pattern shows 97%. This differs markedly from typical classical neural network training where accuracy improves gradually from the start. The sudden transition suggests the quantum circuit explores parameter space initially, finds a productive region, then exploits it for rapid learning. Close alignment between training (blue) and validation (orange) curves indicates excellent generalization without overfitting despite the limited 72-parameter model handling 40,000 training examples.

Metric	VQC	FaceNet-CPU	QSVM
Accuracy	97.0%	85.1%	80.0%
Training Time	0.91 h	243 h	38 h
Face Precision	0.96	0.84	0.82
Face Recall	0.95	0.86	0.85
Face F1-Score	0.955	0.850	0.835
Non-face Precision	0.97	0.86	0.78

Table 1: Comparative Performance on 10,000-Image Test Set

Figure 4: Normalized confusion matrix for VQC validation set with 10,000 test images: 5,000 faces, 5,000 non-faces. The matrix demonstrates strong discrimination capability with minimal confusion between classes.

Class	Precision	Recall	F1-Score	Support
Face	0.96	0.95	0.955	5,000
Non-Face	0.97	0.97	0.970	5,000
Accuracy			0.970	10,000
Macro Avg	0.965	0.960	0.963	10,000
Weighted Avg	0.965	0.960	0.963	10,000

Table 2: Detailed Classification Report on 10,000-Image Test Set

One-time Preprocessing
Operation	Time (s)	Percentage
Face detection (Haar)	127.89	3.9%
ROI extraction/resize	52.41	1.6%
Gamma correction	38.76	1.2%
CLAHE enhancement	74.78	2.3%
Image loading	45.32	1.4%

Table 3: Detailed Timing Analysis (50,000 Images, 100 Epochs)

Figure 5: Comparative performance analysis across three approaches tested on identical hardware (Intel i7, 16GB RAM, CPU-only, no GPU). Green bars (left axis): Classification accuracy showing VQC achieves 97%, outperforming classical FaceNet-CPU (85.1%) and QSVM (80%). Orange bars (right axis, logarithmic scale): Training time in hours demonstrating dramatic computational advantage—VQC trains in 0.91h compared to 243h for classical FaceNet (267× speedup) and 38h for QSVM (42× speedup).The VQC simultaneously achieves higher accuracy and faster training , suggesting genuine quantum advantage for facial biometrics under resource-constrained conditions. Note that GPU-accelerated classical methods would train faster (35-40h) but require specialized hardware unavailable in many practical deployment scenarios.

Figure 6: Statistical stability of the proposed VQC on IBM quantum hardware. Mean confidence scores are shown for different shot budgets, with error bars indicating standard deviation across repeated executions.

Shots	Repetitions	Mean Confidence	Std. Dev.
512	10	0.714	0.057
1024	10	0.696	0.027

Table 4: Hardware Runtime Stability Results on IBM Quantum Backend

Face Recognition

Face recognition matches or identifies people from facial images, videos, and biometric representations across changing conditions.

See all 108 papers

Evolving Hybrid Quantum-Classical Architectures for Image Classification
Oct 2, 2026Devroop Kar, Daniel Krutz, Travis DesellHybrid Quantum-Classical ML Quantum Architecture Search
Rochester Institute of Technology Rochester, NY 14623, USA
A Width-Matched Comparison of Hybrid Quantum-Classical Self-Supervised Learning for Fingerprint Recognition
Sep 30, 2026Maria S. Edwards, Kidwell Dlamini, Pin-An Lin +2Quantum Feature Maps Contrastive Learning
Department of Computer Science and Information Engineering, National Dong Hwa University, Hualien, Taiwan · Department of Computer Science and Information Engineering, National Chiayi University, Chiayi City, Taiwan
Hybrid Classical-Quantum Transfer Learning with Noisy Quantum Circuits
Mar 17, 2026D. Martín-Pérez, F. Rodríguez-Díaz, D. Gutiérrez-Avilés +2Quantum Machine Learning Hybrid Quantum-Classical ML
Data Science and Big Data Lab, Pablo de Olavide University, Seville, Spain · Department of Computer Science, University of Seville, Seville, Spain

Oct 2, 2026J/K move · Enter open · S save

Evolving Hybrid Quantum-Classical Architectures for Image Classification

Devroop Kar, Daniel Krutz, Travis Desell

Rochester Institute of Technology Rochester, NY 14623, USA

Hybrid Quantum-Classical ML Quantum Architecture Search Variational Quantum Circuits Image Classification

Open paper AskPDF arXiv

Hybrid quantum classical neural networks integrate parameterized quantum circuits (PQCs) with established deep learning architectures, but their performance depends strongly on the choice of quantum circuit architecture, a choice that remains largely manual. Most existing approaches rely on hand-designed or fixed circuit ansätze, requiring circuit structure, gate composition, and qubit connectivity to be specified in advance with no guarantee that they suit the task. This limitation is especially acute in image classification, where quantum circuits must transform features extracted by classical networks while remaining compact enough for practical training, requirements that generic, task-agnostic ansätze are unlikely to satisfy simultaneously. We extend EXAQC, an evolutionary framework for automated quantum circuit discovery, to image classification. EXAQC evolves PQCs as intermediate processing modules while retaining classical feature-extraction and prediction layers. On MNIST, Fashion-MNIST, and CIFAR-10, EXAQC achieves 98.42%, 90.62%, and 85.47% accuracy, respectively, while using comparable gate counts to other quantum architecture-search methods. Against classical networks, evolved hybrid models maintain comparable accuracy with substantially fewer trainable parameters, reaching 85.68% on CIFAR-10 with over 25

\times

fewer parameters than a 10-layer CNN. Encoding choice also matters: rotation-based encodings (RX, RY, U3) outperform amplitude encoding by 22-25 points on CIFAR-10. These results demonstrate that automated circuit discovery yields compact quantum modules that can replace larger classical components in vision architectures while retaining competitive accuracy.

Abstract

Face Recognition

Explore similar work

Evolving Hybrid Quantum-Classical Architectures for Image Classification

A Width-Matched Comparison of Hybrid Quantum-Classical Self-Supervised Learning for Fingerprint Recognition

Hybrid Classical-Quantum Transfer Learning with Noisy Quantum Circuits