Abstract We propose a comparative analysis of quantum image classification utilising a circularly-entangled quantum neural network (QNN) in conjunction with four quantum image representation (QIR) strategies: Novel Amplitude Square Sum (NASS), Quantum Block Image Representation (QBIR), Fourier-based Threshold Quantum Representation (FTQR), and Flexible Representation for Quantum Color Image (FRQCI). A methodical assessment of accuracy, loss convergence, and quantum resource requirements was made possible by the encoding of binary MNIST digits (0 and 1) at image sizes of \((2\times2)\), \((4\times4)\), and \((8\times8)\). Our findings demonstrate that, whereas QBIR-QNN offers competitive results with shallow circuits at the expense of greater qubit counts, NASS-QNN continuously converges to high training accuracy with stable parameters, achieving the most reliable and accurate performance. On the other hand, the instability and large resource overheads of FTQR-QNN and FRQCI-QNN restrict their applicability in the NISQ regime. The most promising approaches are highlighted by these results: NASS and QBIR, which provide workable trade-offs between hardware implementability, accuracy, and convergence stability.
Abstract We propose a comparative analysis of quantum image classification utilising a circularly-entangled quantum neural network (QNN) in conjunction with four quantum image representation (QIR) strategies: Novel Amplitude Square Sum (NASS), Quantum Block Image Representation (QBIR), Fourier-based Threshold Quantum Representation (FTQR), and Flexible Representation for Quantum Color Image (FRQCI). A methodical assessment of accuracy, loss convergence, and quantum resource requirements was made possible by the encoding of binary MNIST digits (0 and 1) at image sizes of ((2\times2)), ((4\times4)), and ((8\times8)). Our findings demonstrate that, whereas QBIR-QNN offers competitive results with shallow circuits at the expense of greater qubit counts, NASS-QNN continuously converges to high training accuracy with stable parameters, achieving the most reliable and accurate performance. On the other hand, the instability and large resource overheads of FTQR-QNN and FRQCI-QNN restrict their applicability in the NISQ regime. The most promising approaches are highlighted by these results: NASS and QBIR, which provide workable trade-offs between hardware implementability, accuracy, and convergence stability.
Classification is a core application of quantum machine learning. This study aims at the problems existing in current pure quantum classifiers, such as the large number of consumed qubits and the unsatisfactory performance in multi-classification tasks. Based on the flexible representation of quantum images (FRQI), a new quantum circuit model QCCE that can be used for multi-classification tasks is constructed by introducing additional measurement qubits and connecting them to the color qubit bits. This model not only inherits the encoding advantages of FRQI, enabling it to encode high-dimensional image inputs with a relatively small number of qubits, but also has all trainable parameters in the quantum circuit, with classical computing only providing the loss function and optimizer. Experiments on datasets such as MNIST and Fashion MNIST have shown that in multi-classification tasks like three, four, and five-class classifications, the QCCE model demonstrates obvious advantages over similar quantum circuit models such as HEA and MORE in terms of classification accuracy and convergence speed. The research in this study provides new ideas and methods for the application of quantum machine learning in image processing.
Abstract The Quantum Image Representation (QIR) remains one of the most significant challenges in Quantum Image Processing (QIP) and Quantum Machine Learning (QML) with the resource constraints required by Noisy Intermediate-Scale Quantum (NISQ) hardware. Currently available pixel representation schemes, such as Flexible Representation of Quantum Images (FRQI) and Novel Enhanced Quantum Representation (NEQR), typically have circuit depth and the number of required qubits that grow with image resolution, rendering them impractical for large-scale images. To alleviate these limitations, we propose the Qubit-Efficient and Resolution-Independent Quantum Image Representation (QRISP), a compact, feature-aware encoding framework that is resource-efficient while preserving discriminative information for classification. The proposed approach divides an image into a user-configurable number of non-overlapping blocks, stores complementary statistical descriptors such as mean intensity, local variance, dominant gradient orientation, and calculates the same descriptors using a series of parameterized rotation gates in a quantum feature register; a position register is used to store block indices, and a lightweight entanglement layer is used to store the feature correlations. One of the major features of QRISP is the resolution-independent quantum circuit structure of quantum resources, which depend on the block configuration rather than the image size, allowing flexible processing and trade-offs between representation fidelity and qubit complexity. Experimental results with the MNIST and Fashion-MNIST datasets for the case of Quantum Support Vector Machine (QSVM) and Quantum Neural Network (QNN) pipelines show significant improvement in the number of gates and depth of the circuits, as well as in the time for computing the kernel, without compromising the classification results from FRQI and NEQR, with respect to which their performance is competitive. This shows their potential for scalable, NISQ-oriented quantum image processing.
Noisy Intermediate-Scale Quantum (NISQ) devices have led to the creation of hybrid quantum-classical algorithms, which have provided an innovative design to improve classical machine learning. This article focuses on a review of Hybrid Quantum-Classical Neural Networks (H-QCNNs) in image classification, which is an emerging area that attempts to exploit the representational ability of parameterized quantum circuits (PQCs) and the proven feature extraction capabilities of classical deep learning. We empirically examine the current tendencies in architecture, such as quanvolutional networks, sequential models, and variational quantum classifiers, and critically assess their performance on standard and custom data. The literature analysis shows a large performance difference between noiseless simulation and execution on real hardware, indicating the highly problematic aspects of scaling beyond binary tasks, quantum noise, and a lack of standardized benchmarking. Moreover, we mention such innovative contributions as the small H-QNN model that offers comparable accuracy, yet meets strict NISQ requirements. Last but not least, the current review offers a systematic taxonomy of H-QCNN architectures, a comparative study of methodology options, and a clear roadmap to future research to reach a workable quantum advantage in computer vision. It is a resource to the beginners as well as a reference to the professionals working in the field, so the field would be moved to more powerful, scalable, and hardware-conscious ways of implementation.
Quantum Image Representation (QIR) has become a key component of Quantum Image Processing (QIP), enabling the encoding and manipulation of visual data within quantum computing environments.Its integration with Quantum Artificial Intelligence (QAI) and Quantum Machine Learning (QML) has attracted significant research interest due to the potential advantages of quantum parallelism and compact data representation.This survey reviews major QIR techniques, analyzes their encoding strategies and computational characteristics, and examines their opportunities, challenges, and future relevance in quantum AI and machine learning applications.A systematic review of 70 peer-reviewed journal and conference publications published between 1992 and 2025 was conducted.The selected studies were evaluated based on encoding mechanisms, qubit requirements, circuit complexity, retrieval capability, scalability, and suitability for quantum image processing and learning tasks.The review covers prominent QIR models, including FRQI, NEQR, MCQI, QPIE, GQIR, QRMW, and tensor-network-based representations.Comparative analysis reveals important trade-offs between qubit efficiency, encoding complexity, retrieval accuracy, and scalability.Amplitude-based approaches provide compact image encoding but suffer from probabilistic retrieval, whereas basis-state representations offer deterministic recovery at the expense of increased circuit complexity.QIR demonstrates significant potential in image classification, feature extraction, image compression, and hybrid quantum-classical learning frameworks.QIR provides a promising foundation for next-generation quantum-enhanced visual intelligence systems.However, practical implementation remains constrained by state preparation overhead, noise, decoherence, limited qubit availability, and the lack of standardized evaluation frameworks.Future research should focus on scalable encoding methods, fault-tolerant quantum architectures, benchmarking standards, and deeper integration with advanced QML models to accelerate the development of practical quantum image analytics.
Image classification is an important application for deep learning. With the advent of quantum technology, quantum neural networks (QNNs) have become the focus of research. Traditional deep learning-based image classification involves using a convolutional neural network (CNN) to extract features from the image and a multi-layer perceptron (MLP) network to create the decision boundaries. However, quantum circuits with parameters can extract rich features from images and also create complex decision boundaries. This paper proposes a hybrid QNN (H-QNN) model designed for binary image classification that capitalizes on the strengths of quantum computing and classical neural networks. Our H-QNN model uses a compact, two-qubit quantum circuit integrated with a classical convolutional architecture, making it highly efficient for computation on noisy intermediate-scale quantum (NISQ) devices that are currently leading the way in practical quantum computing applications. Our H-QNN model significantly enhances classification accuracy, achieving a 90.1% accuracy rate on binary image datasets. In addition, we have extensively evaluated baseline CNN and our proposed H-QNN models for image retrieval tasks. The obtained quantitative results exhibit the generalization of our H-QNN for downstream image retrieval tasks. Furthermore, our model addresses the issue of overfitting for small datasets, making it a valuable tool for practical applications.
Image classification is an important task in many practical applications. Recently, there have been various works on classification problems using quantum neural network (QNN). Different data encoders have been investigated, but efficient quantum encoding schemes for high-dimensional data such as images are still an open question. In this study, we investigate popular representation formats of quantum images, namely FRQI and NEQR, as data encoders for quantum neural networks. It is found that the FRQI format not only provides the best classification accuracy compared to NEQR and other popular encoding methods but also is consistent across different circuit depths of QNN. It should be noted that FRQI is notorious for its probabilistic representation in quantum image processing. The finding of this study is interesting because FRQI can play an important role in the field of quantum machine learning.