Low-resolution Images Research Articles

A novel facial age estimation through super-resolution based on deep learning

Age estimation encounters challenges due to the low resolution (LR) of images captured in real-world scenarios, emphasizing the critical role of image quality in achieving accurate age estimations. Despite efforts to enhance age estimation through LR image training, existing methodologies often overlook the incorporation of super-resolution (SR) as a crucial pre-processing step for LR images. This study introduces an innovative approach that integrates SR of facial images with age estimation methodologies. Specifically, a hierarchical reconstruction network (HRN) structure is employed to train LR images, fully integrating shallow concrete features and deep abstract features. Validation of the training outcomes includes evaluation metrics such as peak signal-to-noise ratio and structural similarity index measurement, showing superior performance compared to several state-of-the-art approaches. Subsequently, an enhanced soft stage regression network is designed for facial age measurement and trained using the SR-IMDB dataset generated by HRN to minimize mean absolute error. This approach achieves a final result of 7.33, indicating an average performance improvement of 10.94% compared to five other state-of-the-art methods. Experimental results highlight the effectiveness of integrating SR into the age estimation process.

Movie Character Recognition from Video and Image using AI

Recognition of movie characters using video and images An innovative research study called Using AI explores the use of Artificial Intelligence (AI) for recognizing and identifying movie characters using still photos and video clips. The study offers a new AI system that analyses visual input, extracts attributes, and correctly identifies movie characters using machine learning and deep learning methods. This system is able to identify characters in a single frame and follow their actions over several frames, giving it a thorough understanding of the character's function in the story. The technology's potential uses in a number of industries, including film creation, advertising, and audience analytics. The suggested AI system can automate video editing, help with character recognition-based targeted advertising, and offer insightful data on audience preferences and habits. The implications of this technology are enormous, creating new opportunities for the film industry to grow technologically. The challenges and limitations of implementing AI in movie character recognition. It highlights issues such as handling low-resolution images, varying lighting conditions, and changes in character appearance over time. Despite these challenges, the AI system demonstrates a high degree of accuracy and efficiency. The concludes by suggesting future enhancements to the system.

Impact of PET Reconstruction on Amyloid-β Quantitation in Cross-Sectional and Longitudinal Analyses.

Amyloid-β (Aβ) accumulation in Alzheimer disease (AD) is typically measured using SUV ratio and the centiloid (CL) scale. The low spatial resolution of PET images is known to degrade quantitative metrics because of the partial-volume effect. This article examines the impact of spatial resolution, as determined by the reconstruction configuration, on the Aβ PET quantitation in both cross-sectional and longitudinal data. Methods: The cross-sectional study involved 89 subjects with 20-min [18F]florbetapir scans generated on an mCT (44 Aβ-negative [Aβ-], 45 Aβ-positive [Aβ+]) using 69 reconstruction configurations, which varied in number of iteration updates, point-spread function, time-of-flight, and postreconstruction smoothing. The subjects were classified as Aβ- or Aβ+ visually. For each reconstruction, Aβ CL was calculated using CapAIBL, and the spatial resolution was calculated as full width at half maximum (FWHM) using the barrel phantom method. The change in CLs and the effect size of the difference in CLs between Aβ- and Aβ+ groups with FWHM were examined. The longitudinal study involved 79 subjects (46 Aβ-, 33 Aβ+) with three 20-min [18F]flutemetamol scans generated on an mCT. The subjects were classified as Aβ- or Aβ+ using a cutoff CL of 20. All scans were reconstructed using low-, medium-, and high-resolution configurations, and Aβ CLs were calculated using CapAIBL. Since linear Aβ accumulation was assumed over a 10-y interval, for each reconstruction configuration, Aβ accumulation rate differences (ARDs) between the second and first periods were calculated for all subjects. Zero ARD was used as a consistency metric. The number of Aβ accumulators was also used to compare the sensitivity of CL across reconstruction configurations. Results: In the cross-sectional study, CLs in both the Aβ- and the Aβ+ groups were impacted by the FWHM of the reconstruction method. Without postreconstruction smoothing, Aβ- CLs increased for a FWHM of 4.5 mm or more, whereas Aβ+ CLs decreased across the FWHM range. High-resolution reconstructions provided the best statistical separation between groups. In the longitudinal study, the median ARD of low-resolution reconstructed data for the Aβ- group was greater than zero whereas the ARDs of higher-resolution reconstructions were not significantly different from zero, indicating more consistent rate estimates in the higher-resolution reconstructions. Higher-resolution reconstructions identified 10 additional Aβ accumulators in the Aβ- group, resulting in a 22% increased group size compared with the low-resolution reconstructions. Higher-resolution reconstructions reduced the average CLs of the negative group by 12 points. Conclusion: High-resolution PET reconstructions, inherently less impacted by partial-volume effect, may improve Aβ PET quantitation in both cross-sectional and longitudinal data. In the cross-sectional analysis, separation of CLs between Aβ- and Aβ+ cohorts increased with spatial resolution. Higher-resolution reconstructions also exhibited both improved consistency and improved sensitivity in measures of Aβ accumulation. These features suggest that higher-resolution reconstructions may be advantageous in early-stage AD therapies.

LED-based temporal variant noise model for Fourier ptychographic microscopy

Fourier ptychographic microscopy (FPM) is a technique to reconstruct a high-resolution image from a set of low-resolution images captured with different illumination angles, which is susceptible to ambient noise, system noise, and weak currents when acquiring large-angle images, especially dark field images. To effectively address the noise problem, we propose an adaptive denoising algorithm based on a LED-based temporal variant noise model. Taking the results of blank slide samples as the reference value of noise, and analyzing the distribution of noise, we establish a statistical model for temporal variant noise, describing the relationship between temporal noise and LED spatial location. Based on this model, Gaussian denoising parameters are selected to adaptively denoise the images with different locations, with which high-resolution images can be reconstructed. Compared with other methods, the experimental results show that the proposed method effectively suppresses the noise, recovers more image details, increases the image contrast, and obtains better visual effects. Meanwhile, better objective evaluation also mirrors the advantages of the proposed algorithms.

Utilizing EfficientNet for sheep breed identification in low-resolution images

Automatically recognizing sheep breeds is highly valuable for the sheep farming industry, allowing farmers to pinpoint their specific business needs. Accurately distinguishing between sheep breeds poses a challenge for numerous farmers with limited expertise. Although biometric-based identification offers a feasible solution, its application becomes impractical when assessing large numbers of sheep in real-time. Therefore, the implementation of an automatic sheep classification model that can replicate the breed identification skills of a sheep breed expert can come in handy. This would be particularly beneficial for novice farmers who could utilize handheld devices for breed classification. To address this objective, we propose employing a convolutional neural network (CNN) model capable of rapidly and accurately identifying sheep breeds from low-resolution images. Our experiment utilized a dataset of 1680 facial images representing four distinct sheep breeds. We conducted experiments on the dataset using various EfficientNet models and found that EfficientNetB5 achieved the highest performance with 97.62 % accuracy on a 10 % test split. The classification model we developed has the potential to assist sheep farmers in efficiently distinguishing between different breeds, facilitating more precise assessments and sector-specific classification for various businesses within the industry.

A hyperspectral surgical microscope with super-resolution reconstruction for intraoperative image guidance.

Hyperspectral imaging (HSI) is an emerging imaging modality in medical applications, especially for intraoperative image guidance. A surgical microscope improves surgeons' visualization with fine details during surgery. The combination of HSI and surgical microscope can provide a powerful tool for surgical guidance. However, to acquire high-resolution hyperspectral images, the long integration time and large image file size can be a burden for intraoperative applications. Super-resolution reconstruction allows acquisition of low-resolution hyperspectral images and generates high-resolution HSI. In this work, we developed a hyperspectral surgical microscope and employed our unsupervised super-resolution neural network, which generated high-resolution hyperspectral images with fine textures and spectral characteristics of tissues. The proposed method can reduce the acquisition time and save storage space taken up by hyperspectral images without compromising image quality, which will facilitate the adaptation of hyperspectral imaging technology in intraoperative image guidance.

Accurate and safe diagnosis and treatment of neoplastic biliary lesions using a novel 9F and 11F digital single-operator cholangioscope.

Background and study aims Digital single-operator cholangioscopy (DSOC) allows the diagnosis of biliary duct disorders and treatment for complicated stones. However, these technologies have limitations such as the size of the probe and working channel, excessive cost, and low image resolution. Recently, a novel DSOC system (eyeMAX, Micro-Tech, Nanjing, China) was developed to address these limitations. We aimed to evaluate the usefulness and safety of a novel 9F and 11F DSOC system in terms of neoplastic diagnostic accuracy based on visual examination, ability to evaluate tumor extension and to achieve complete biliary stone clearance, and procedure-related adverse events (AEs). Patients and methods Data from ≥ 18-year-old patients who underwent DSOC from July 2021 to April 2022 were retrospectively recovered and divided into a diagnostic and a therapeutic cohort. Results A total of 80 patients were included. In the diagnostic cohort (n = 49/80), neovascularity was identified in 26 of 49 patients (46.9%). Biopsy was performed in 65.3% patients with adequate tissue sample obtained in 96.8% of cases. Biopsy confirmed neoplasia in 23 of 32 cases. DSOC visual impression achieved 91.6% sensitivity and 87.5% specificity in diagnosing neoplasms. In the therapeutic cohort (n = 43/80), 26 of 43 patients required lithotripsy alone. Total stone removal was achieved in 71% patients in the first session. Neither early nor late AEs were documented in either the diagnostic or therapeutic cohort. Conclusions The novel DSOC device has excellent diagnostic accuracy in distinguishing neoplastic biliary lesions as well as therapeutic benefits in the context of total stone removal, with no documented AEs.

Locality-constraint Representation with Minkowski distance metric for an effective Face Hallucination

Face hallucination (FaceH), the domain specific super resolution technique has significant importance in the field of image processing and computer vision. These methods provide visually convincing high-resolution face images using the low-resolution input images. To do this, all the state-of-the-art methods make use of distance measure algorithms. The performance of distance measure is one of the core components that impact the performance of FaceH. However, there are many k-NN distance measure algorithms, to the best of our knowledge, the most commonly used metric is the Euclidean distance, though it is the best, a little deviation may result in large elevation on estimating the reconstructed weight of the pixel. To overcome this, it is necessary to find an alternative method and this motivated to develop Minkowski Locality-constrained Representation (MinLcR) FaceH that makes use of Minkowski distance and after which it is integrated with Locality-constrained Representation (LcR) that incorporates the contextual information of the neighboring pixel for better hallucination. The main idea of MinLcR is to make a difference in the distance metric and compare it with methods that make use of Euclidean distance. Experimental analysis carried out using FEI database shows that the proposed method has an effective improvement over other methods.

Detection of Dense Built-Up Area in Low-Resolution Satellite Images Using Deep Learning and DBSCAN Approaches

Recent developments in satellite image processing tend to eliminate the need for intensive on-site surveys of urban or rural areas for infrastructure allocation planning. In particular, the detection of buildings in satellite images can significantly aid in rural or urban planning. However, detecting individual buildings in low-resolution satellite images is challenging due to a lack of visual clarity. In order to address this problem, we propose a computer vision-based hybrid framework to detect densely constructed building regions in low resolution satellite images, which can also serve as an assistive framework for automated geo-spatial survey of various sites. Our hybrid framework is comprised of three modules, namely the Mask R-CNN module, an ANN-based refinement module, and a DBSCAN based refinement module. The Mask R-CNN is employed to predict probable clustered building regions in the satellite image, whereas the ANN-based refinement module is applied to remove homogeneous regions (e.g. vegetation area) from the Mask R-CNN-detected clustered building region. Finally, a DBSCAN-based refinement module removes non-congested built-up regions so that densely constructed built-up areas can be identified with better precision. Our experimentation using publicly accessible satellite image datasets (AIRS dataset-Kaggle) establishes the effectiveness of the proposed hybrid framework in identifying densely populated building areas in low-resolution satellite images. The outcomes of the proposed framework demonstrated an approximately 5–10% improvement over various Mask R-CNN and U-Net-based approaches in terms of achieving better precision value of the detection.

Quantifying ecosystem services from trees by using i-tree with low-resolution satellite images

Air pollution on a worldwide scale poses significant risks to our health and food security. Trees, as ecosystem units, play a huge role in mitigating air pollution but quantifying this ecosystem service in the most polluted regions of the world is difficult due to a lack of relevant data. The ability of trees to mitigate air emissions is evaluated by the i-Tree Eco tool, which uses the tree inventory data based on high-resolution data such as LiDAR, which is predominantly available for the countries in the global north. In contrast, for most countries in the global south, collating such information is difficult and, therefore, obtaining the tree inventory and usage of i-Tree Eco is challenging. To circumvent these lacunae, in this study, we propose a method hinging on remote sensing based on readily available Landsat satellite images. We validate our approach on Balboa Park, San Diego, California, for which both LiDAR and Landsat data are available, and showcase its applicability to the South Delhi and Chennai city regions of India, where LiDAR data are not available.

MiSleep: Human Sleep Posture Identification from Deep Learning Augmented Millimeter-wave Wireless Systems

In this work, we propose MiSleep , a deep learning augmented millimeter-wave (mmWave) wireless system to monitor human sleep posture by predicting the 3D location of the body joints of a person during sleep. Unlike existing vision- or wearable-based sleep monitoring systems, MiSleep is not privacy-invasive and does not require users to wear anything on their body. MiSleep leverages knowledge of human anatomical features and deep learning models to solve challenges in existing mmWave devices with low-resolution and aliased imaging and specularity in signals. MiSleep builds the model by learning the relationship between mmWave reflected signals and body postures from thousands of existing samples. Since a practical sleep also involves sudden toss-turns, which could introduce errors in posture prediction, MiSleep designs a state machine based on the reflected signals to classify the sleeping states into rest or toss-turn and predict the posture only during the rest states. We evaluate MiSleep with real data collected from Commercial-Off-The-Shelf mmWave devices for eight volunteers of diverse ages, genders, and heights performing different sleep postures . We observe that MiSleep identifies the toss-turn events start time and duration within 1.25 s and 1.7 s of the ground truth, respectively, and predicts the 3D location of body joints with a median error of 1.3 cm only and can perform even under the blankets, with accuracy on par with the existing vision-based system, unlocking the potential of mmWave systems for privacy-noninvasive at-home healthcare applications.

Enhancing Cable Vibration Measurement at Long Distances Through Super-Resolution Reconstruction and Target Foreground Segmentation

This paper presents a comprehensive non-contact computer vision-based system for monitoring cable vibrations in cable-supported bridges, addressing challenges related to low image resolution and feature extraction difficulties at long distances. The proposed system utilizes deep learning techniques to enhance cable vibration recognition accuracy and offers a practical solution for cable monitoring without the need for target assistance. The core of the system is a novel two-stage model, which combines a super-resolution (SR) video reconstruction algorithm with state-of-the-art Resnet-34 and Swin-B models for precise target foreground segmentation. This approach significantly improves the recognition of target details and enhances the accuracy of cable vibration data in monitoring videos. Furthermore, a phase-based motion estimation (PME) algorithm is employed for precise cable vibration measurement. Field tests conducted on two cable-supported bridges validate the effectiveness of the system. The results demonstrate superior accuracy and noise immunity compared to traditional methods, achieving sub-pixel level precision with a maximum error rate below 2%. This system represents a significant advancement in non-contact structural health monitoring for long-distance cable vibration monitoring in cable-supported bridges.

A new degradation model and an improved SRGAN for multi-image super-resolution reconstruction

ABSTRACT In order to solve the problems existing in multi-image super-resolution reconstruction methods, such as the difficulty of acquiring and processing multiple low-resolution images, the inability to make full use of the complementary information between different images, and the loss of details, this paper proposes a new degradation model and an improved SRGAN for multi-image super-resolution reconstruction. Firstly, a new degradation model considering both masked autoencoding and downsampling (DMMD) is designed, which can simulate the complex degradation environment in real life well and reduce the difficulty of obtaining and processing multiple low-resolution images. Then, we design a weight setting strategy for image fusion to make full use of the complementary information between different low-resolution images. To enhance the network's attention and propagation of high-frequency effective information in the feature map, the convolution block attention module (CBAM) is introduced into the generator network of SRGAN model, and a SRGAN combined with CBAM (SRGANCBAM) is designed. Finally, the fused low-resolution image is input into SRGANCBAM for reconstruction to obtain its corresponding high-resolution image. The experimental results show that our DMMD can solve the problem that it is difficult to acquire and process multiple low-resolution images. The complementary information between different images is fully utilized by the reasonable weight setting strategy. On four public datasets, the PSNR value of our SRGANCBAM can be 0.532, 0.207, 0.357 and 0.537 higher than the baseline model, respectively. Compared with the state-of-the-art methods, our SRGANCBAM achieves higher values of evaluation metrics and reconstructs clearer and more realistic images.

PIS-Net: Efficient Medical Image Segmentation Network with Multivariate Downsampling for Point-of-Care.

Recently, with more portable diagnostic devices being moved to people anywhere, point-of-care (PoC) imaging has become more convenient and more popular than the traditional "bed imaging". Instant image segmentation, as an important technology of computer vision, is receiving more and more attention in PoC diagnosis. However, the image distortion caused by image preprocessing and the low resolution of medical images extracted by PoC devices are urgent problems that need to be solved. Moreover, more efficient feature representation is necessary in the design of instant image segmentation. In this paper, a new feature representation considering the relationships among local features with minimal parameters and a lower computational complexity is proposed. Since a feature window sliding along a diagonal can capture more pluralistic features, a Diagonal-Axial Multi-Layer Perceptron is designed to obtain the global correlation among local features for a more comprehensive feature representation. Additionally, a new multi-scale feature fusion is proposed to integrate nonlinear features with linear ones to obtain a more precise feature representation. Richer features are figured out. In order to improve the generalization of the models, a dynamic residual spatial pyramid pooling based on various receptive fields is constructed according to different sizes of images, which alleviates the influence of image distortion. The experimental results show that the proposed strategy has better performance on instant image segmentation. Notably, it yields an average improvement of 1.31% in Dice than existing strategies on the BUSI, ISIC2018 and MoNuSeg datasets.

Deep local-to-global feature learning for medical image super-resolution

Medical images play a vital role in medical analysis by providing crucial information about patients’ pathological conditions. However, the quality of these images can be compromised by many factors, such as limited resolution of the instruments, artifacts caused by movements, and the complexity of the scanned areas. As a result, low-resolution (LR) images cannot provide sufficient information for diagnosis. To address this issue, researchers have attempted to apply image super-resolution (SR) techniques to restore the high-resolution (HR) images from their LR counterparts. However, these techniques are designed for generic images, and thus suffer from many challenges unique to medical images. An obvious one is the diversity of the scanned objects; for example, the organs, tissues, and vessels typically appear in different sizes and shapes, and are thus hard to restore with standard convolution neural networks (CNNs). In this paper, we develop a dynamic-local learning framework to capture the details of these diverse areas, consisting of deformable convolutions with adjustable kernel shapes. Moreover, the global information between the tissues and organs is vital for medical diagnosis. To preserve global information, we propose pixel–pixel and patch–patch global learning using a non-local mechanism and a vision transformer (ViT), respectively. The result is a novel CNN-ViT neural network with Local-to-Global feature learning for medical image SR, referred to as LGSR, which can accurately restore both local details and global information. We evaluate our method on six public datasets and one large-scale private dataset, which include five different types of medical images (i.e., Ultrasound, OCT, Endoscope, CT, and MRI images). Experiments show that the proposed method achieves superior PSNR/SSIM and visual performance than the state of the arts with competitive computational costs, measured in network parameters, runtime, and FLOPs. What is more, the experiment conducted on OCT image segmentation for the downstream task demonstrates a significantly positive performance effect of LGSR.

Efficient Conditional Diffusion Model with Probability Flow Sampling for Image Super-resolution

Image super-resolution is a fundamentally ill-posed problem because multiple valid high-resolution images exist for one low-resolution image. Super-resolution methods based on diffusion probabilistic models can deal with the ill-posed nature by learning the distribution of high-resolution images conditioned on low-resolution images, avoiding the problem of blurry images in PSNR-oriented methods. However, existing diffusion-based super-resolution methods have high time consumption with the use of iterative sampling, while the quality and consistency of generated images are less than ideal due to problems like color shifting. In this paper, we propose Efficient Conditional Diffusion Model with Probability Flow Sampling (ECDP) for image super-resolution. To reduce the time consumption, we design a continuous-time conditional diffusion model for image super-resolution, which enables the use of probability flow sampling for efficient generation. Additionally, to improve the consistency of generated images, we propose a hybrid parametrization for the denoiser network, which interpolates between the data-predicting parametrization and the noise-predicting parametrization for different noise scales. Moreover, we design an image quality loss as a complement to the score matching loss of diffusion models, further improving the consistency and quality of super-resolution. Extensive experiments on DIV2K, ImageNet, and CelebA demonstrate that our method achieves higher super-resolution quality than existing diffusion-based image super-resolution methods while having lower time consumption. Our code is available at https://github.com/Yuan-Yutao/ECDP.

LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)

Recent advances in image-to-image translation involve the integration of non-visual imagery in deep models. Non-visual sensors, although more costly, often produce low-resolution images. To combat this, methods using RGB images to enhance the resolution of these modalities have been introduced. Fusing these modalities to achieve high-resolution results demands models with millions of parameters and extended inference times. We present LaMAR, a lightweight model. It employs Laplacian image pyramids combined with a low-resolution thermal image for Guided Thermal Super Resolution. By decomposing the RGB image into a Laplacian pyramid, LaMAR preserves image details and avoids high-resolution feature map computations, ensuring efficiency. With faster inference times and fewer parameters, our model demonstrates state-of-the-art results.

Unveiling Details in the Dark: Simultaneous Brightening and Zooming for Low-Light Image Enhancement

Existing super-resolution methods exhibit limitations when applied to nighttime scenes, primarily due to their lack of adaptation to low-pair dynamic range and noise-heavy dark-light images. In response, this research introduces an innovative customized framework to simultaneously Brighten and Zoom in low-resolution images captured in low-light conditions, dubbed BrZoNet. The core method begins by feeding low-light, low-resolution images, and their corresponding ground truths into the Retinex-induced siamese decoupling network. This process yields distinct reflectance maps and illuminance maps, guided by supervision from the ground truth’s decomposition maps. Subsequently, these reflectance and illuminance maps transition into an intricate super-resolution sub-network. This sub-network employs a meticulously designed cross-layer content-aware interactor - Illumination-aware Interaction Unit(IaIU), elegantly endowed with a gating mechanism. The IaIU facilitates meaningful feature interaction between illuminance and reflectance features while effectively reducing unwanted noise. An intricate super-resolution cage is also constructed to comprehensively integrate information, ultimately resulting in the generation of high-resolution images featuring intricate details. Thorough and diverse experiments validate the superiority of the proposed BrZoNet, surpassing contemporary cutting-edge technologies by proficiently augmenting brightness and intricately recovering complex details, showcasing advancements of 7.1% in PSNR, 2.4% in SSIM, and an impressive 36.8% in LPIPS metrics.

MFTN: Multi-Level Feature Transfer Network Based on MRI-Transformer for MR Image Super-resolution

Due to the unique environment and inherent properties of magnetic resonance imaging (MRI) instruments, MR images typically have lower resolution. Therefore, improving the resolution of MR images is beneficial for assisting doctors in diagnosing the condition. Currently, the existing MR image super-resolution (SR) methods still have the problem of insufficient detail reconstruction. To overcome this issue, this paper proposes a multi-level feature transfer network (MFTN) based on MRI-Transformer to realize SR of low-resolution MRI data. MFTN consists of a multi-scale feature reconstruction network (MFRN) and a multi-level feature extraction branch (MFEB). MFRN is constructed as a pyramid structure to gradually reconstruct image features at different scales by integrating the features obtained from MFEB, and MFEB is constructed to provide detail information at different scales for low resolution MR image SR reconstruction by constructing multiple MRI-Transformer modules. Each MRI-Transformer module is designed to learn the transfer features from the reference image by establishing feature correlations between the reference image and low-resolution MR image. In addition, a contrast learning constraint item is added to the loss function to enhance the texture details of the SR image. A large number of experiments show that our network can effectively reconstruct high-quality MR Images and achieves better performance compared to some state-of-the-art methods. The source code of this work will be released on GitHub.

KeDuSR: Real-World Dual-Lens Super-Resolution via Kernel-Free Matching

Dual-lens super-resolution (SR) is a practical scenario for reference (Ref) based SR by utilizing the telephoto image (Ref) to assist the super-resolution of the low-resolution wide-angle image (LR input). Different from general RefSR, the Ref in dual-lens SR only covers the overlapped field of view (FoV) area. However, current dual-lens SR methods rarely utilize these specific characteristics and directly perform dense matching between the LR input and Ref. Due to the resolution gap between LR and Ref, the matching may miss the best-matched candidate and destroy the consistent structures in the overlapped FoV area. Different from them, we propose to first align the Ref with the center region (namely the overlapped FoV area) of the LR input by combining global warping and local warping to make the aligned Ref be sharp and consistent. Then, we formulate the aligned Ref and LR center as value-key pairs, and the corner region of the LR is formulated as queries. In this way, we propose a kernel-free matching strategy by matching between the LR-corner (query) and LR-center (key) regions, and the corresponding aligned Ref (value) can be warped to the corner region of the target. Our kernel-free matching strategy avoids the resolution gap between LR and Ref, which makes our network have better generalization ability. In addition, we construct a DuSR-Real dataset with (LR, Ref, HR) triples, where the LR and HR are well aligned. Experiments on three datasets demonstrate that our method outperforms the second-best method by a large margin. Our code and dataset are available at https://github.com/ZifanCui/KeDuSR.