QUIET-SR: Quantum Image Enhancement Transformer for Single Image Super-Resolution

Siddhant Dutta, Nouhaila Innan, Khadijeh Najafi, Sadok Ben Yahia, Muhammad Shafique · 2025

Read QUIET-SR: Quantum Image Enhancement Transformer for Single Image Super-Resolution by Siddhant Dutta, Nouhaila Innan, Khadijeh Najafi, Sadok Ben Yahia, Muhammad Shafique (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-08759.txt.

Introduction

Siddhant Dutta and four co-authors sharpen low-resolution images with a hybrid quantum model in a 2025 paper. Classic upscaling networks work well but carry heavy parameter counts. Quantum methods promise compact power, yet scaling them to real images remains difficult on today's noisy hardware.

Their framework extends a shifted-window transformer with a quantum attention step. The quantum part runs on small circuits suited to current devices. It helps the model learn the mapping from blurry inputs to crisp outputs. Tests on handwritten digits, fashion images, and medical sets report strong quality scores. The authors frame the design as a practical bridge between quantum methods and everyday image repair tasks.

Catalog record

CREATORSiddhant Dutta, Nouhaila Innan, Khadijeh Najafi, Sadok Ben Yahia, Muhammad Shafique
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-08759

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

QUIET-SR: Quantum Image Enhancement Transformer for Single Image Super-Resolution Siddhant Dutta, Nouhaila Innan, Khadijeh Najafi, Sadok Ben Yahia, Muhammad Shafique Abstract Recent advancements in Single-Image Super-Resolution (SISR) using deep learning have significantly improved image restoration quality. However, the high computational cost of processing high-resolution images due to the large number of parameters in classical models, along with the scalability challenges of quantum algorithms for image processing, remains a major obstacle. In this paper, we propose the Quantum Image Enhancement Transformer for Super-Resolution (QUIET-SR), a hybrid framework that extends the Swin transformer architecture with a novel shifted quantum window attention mechanism, built upon variational quantum neural networks. QUIET-SR effectively captures complex residual mappings between low-resolution and high-resolution images, leveraging quantum attention mechanisms to enhance feature extraction and image restoration while requiring a minimal number of qubits, making it suitable for the Noisy Intermediate-Scale Quantum (NISQ) era. We evaluate our framework in MNIST (30.24 PSNR, 0.989 SSIM), FashionMNIST (29.76 PSNR, 0.976 SSIM) and the MedMNIST dataset collection, demonstrating that QUIET-SR achieves PSNR and SSIM scores comparable to state-of-the-art methods while using fewer parameters. Our efficient batching strategy directly enables massive parallelization on multiple QPU’s paving the way for practical quantum-enhanced image super-resolution through coordinated QPU–GPU quantum supercomputing. I Introduction Single-Image Super-Resolution (SISR) aims to recover a High-Resolution (HR) image from a Low-Resolution (LR) input image [ 4 ] . Formally, given a low-resolution image I_{LR} , the objective of SISR is to reconstruct a high-resolution image I_{HR} such that the perceptual quality and structural fidelity of the resulting image closely approximate those of an ideal high-resolution reference. More formally, the original high-resolution image can be considered the ground truth, representing the scene sampled at a sufficiently high spatial resolution to capture fine details without degradation. In a physical sense, real-world scenes exist as continuous signals, and an ideal high-resolution image corresponds to a discretized representation with minimal loss of spatial information. The aim is to learn a mapping f:I_{LR}\rightarrow I_{HR} that reconstructs an image I_{HR} approximating this ideal ground truth. This technique has significant applications in fields such as medical imaging, satellite imagery, and autonomous driving as it helps enhance scans like MRIs, aiding in disease detection, improves the clarity of earth observation data for environmental monitoring and disaster response, and sharpens camera inputs, improving object recognition for safer navigation respectively. By recovering lost details, SISR enables better decision-making across these fields [ 29 , 15 , 22 ] . In the context of deep learning, traditional SISR methods often use Convolutional Neural Networks (CNNs) that incorporate residual learning, a technique that helps the network focus on reconstructing lost high-frequency details rather than relearning the entire image structure [ 9 , 20 ] . Another common approach involves Generative Adversarial Networks (GANs), which generate high-resolution images by training two competing networks to improve image quality [ 18 ] . However, these methods are computationally demanding and struggle to scale efficiently as the size of the image and complexity increase. Recent developments in SISR models, such as those based on Swin Transformer architectures, have demonstrated state-of-the-art performance in capturing fine details while maintaining contextual integrity, highlighting the ongoing demand for scalable and efficient approaches in this field [ 19 , 7 ] . Figure 1: This grid compares low-resolution (14×14), high-resolution (28×28), and QUIET-SR super-resolution images across multiple MNIST-like datasets. The high-resolution row represents the ground truth, while the QUIET-SR row demonstrates the model’s capability to reconstruct fine details and preserve structural integrity. The super-resolved images generated by QUIET-SR closely approximate the high-resolution ground truth, effectively enhancing image clarity and preserving key features. Quantum Computing (QC) introduces a representational-based computational paradigm that leverages quantum states to encode and manipulate high-dimensional data more efficiently. This approach has the potential to overcome certain limitations of classical computation in tasks requiring complex feature representations, such as extracting fine textures, spatial relationships, and high-frequency details in image processing [ 30 ] . Classical methods often struggle with the exponential growth of data, high memory and computational demands, and the difficulty of solving non-convex optimization problems, making quantum techniques a promising alternative for more scalable and efficient solutions. The field of Quantum Machine Learning (QML) utilizes the unique properties of quantum mechanics to investigate state spaces that are infeasible via classical approaches [ 14 , 3 ] , to improve learning efficiency and model generalization. However, practical implementations of quantum image processing face challenges, primarily due to limitations in qubit resources and the presence of quantum noise [ 30 , 24 ] . Current research in quantum Super-Resolution (SR) predominantly focuses on theoretical frameworks with limited practical implementations. This is largely due to the qubit requirements that exceed the capabilities of existing hardware [ 28 , 6 ] . Several works have employed adiabatic QC with D-Wave systems, achieving proof-of-concept results [ 5 ] . In this paper, we tackle the dual challenge of leveraging Noisy Intermediate-Scale Quantum (NISQ) computers [ 23 ] while efficiently integrating QC with classical deep learning for super-resolution. A major obstacle is the inherent difficulty of combining quantum feature representations with classical deep learning models in a way that effectively utilizes the potential quantum advantages without being bottlenecked by hardware limitations. This challenge motivates our exploration of NISQ devices and the development of hybrid approaches that can overcome their constraints while achieving comparable or improved performance in SISR. To address this, we introduce a novel hybrid quantum-classical architecture for SISR. Our proof-of-concept framework demonstrates a scalable hybrid quantum super-resolution system, operating with fewer than 10 qubits per circuit within the proposed architecture. This represents a step toward practical quantum applications in image processing while staying within current quantum hardware constraints, as shown in Fig. 1 . Our work as shown 2 presents the 1 st resource-aware quantum attention module for image SR, preserving Swin-style shifted-window locality & running on today’s \leq 10-qubit devices. Prior quantum vision approaches either (i) use non-scalable encodings, (ii) address only classification, or (iii) rely on quantum annealing. In contrast, our gate-based, NISQ-compatible design directly tackles SR, marking the 1 st quantum realization of shifted-window attention. Figure 2: High-level workflow of the QUIET-SR framework. The architecture processes a 14\times 14 Low-Resolution (LR) input through three main stages: (1) Shallow Feature Extraction , where initial convolutional layers capture low-frequency information such as edges, textures, and colors; (2) Deep Feature Extraction , which utilizes Quantum Residual Transformer Blocks (Quantum RTSB) containing the specific SQWIN (Shifted Quantum Window) attention mechanism to model complex dependencies; and (3) High Quality Image Reconstruction , where features are aggregated via global residual connections surmised via (4-5) upsampling via a Pixel Shuffle operation and final convolutions to synthesize the 28\times 28 Super-Resolved output. (6) The network is optimized by minimizing the \mathcal{L}_{1} loss between the generated image and the High-Resolution ground truth. The QNN architecture uses angle embedding and basic entangler layers to transform features and optimize attention efficiency. The final stage, SR Image Reconstruction, synthesizes the SR output, demonstrating the advantages of quantum-enhanced image restoration Significance for the CV community: SQWIN, as a technique is the 1 st scalable NISQ compatible method that bridges QC & CV. It uses quantum-native shifted window attention to enable entangled

Read the complete text