Article
Computer Science, Information Systems
Ji-Soo Kim, Keunsoo Ko, Hanul Kim, Chang-Su Kim
Summary: In this paper, a progressive face super-resolution network called RPF is proposed to generate high-resolution facial images without losing details and personal identity. The network utilizes a high-resolution reference image of the same person as the input to preserve both details and identity information.
Article
Computer Science, Artificial Intelligence
Cheng Ma, Yongming Rao, Jiwen Lu, Jie Zhou
Summary: Structures play a crucial role in single image super-resolution (SISR), and this paper proposes a structure-preserving super-resolution (SPSR) method to address the issue of structural distortions. By utilizing gradient guidance and a learnable neural structure extractor (NSE), our method achieves superior results in both detail recovery and structure preservation.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Kelvin C. K. Chan, Xiangyu Xu, Xintao Wang, Jinwei Gu, Chen Change Loy
Summary: In this study, we propose a method called GLEAN for improving the performance of image super-resolution using pre-trained Generative Adversarial Networks (GANs) such as StyleGAN and BigGAN. Unlike existing perceptual-oriented approaches, GLEAN directly leverages the rich and diverse priors encapsulated in a pre-trained GAN for generating realistic outputs. Our approach only requires a single forward pass for restoration, avoiding expensive image-specific optimization at runtime. We also introduce a lightweight version of GLEAN, named LightGLEAN, which achieves comparable image quality with significantly fewer parameters and FLOPs.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Radiology, Nuclear Medicine & Medical Imaging
Keewon Shin, Jung Su Lee, Ji Young Lee, Hyunsu Lee, Jeongseok Kim, Jeong-Sik Byeon, Hwoon-Yong Jung, Do Hoon Kim, Namkug Kim
Summary: Generative adversarial networks (GAN) in medicine are valuable techniques for producing high-quality gastrointestinal images. This study used the progressive growing of GAN (PGGAN) to generate realistic images and investigated its limitations. The accuracy and sensitivity of endoscopists in distinguishing real and synthetic images were not significantly different. Real images with the anatomical landmark pylorus had higher detection sensitivity. However, GANs need improvement in representing rugal folds and mucous membrane texture.
JOURNAL OF DIGITAL IMAGING
(2023)
Article
Computer Science, Artificial Intelligence
Peng Yi, Zhongyuan Wang, Kui Jiang, Junjun Jiang, Tao Lu, Jiayi Ma
Summary: This paper proposes a progressive fusion network for video super-resolution, which effectively utilizes temporal information through progressive separation and fusion of frames. The network incorporates multi-scale structure, hybrid convolutions, and non-local operation to capture a wide range of dependencies and replace traditional motion estimation and compensation. The improved generative adversarial training method avoids temporal artifacts and generates more realistic and temporally consistent videos.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Jin Zhu, Chuan Tan, Junwei Yang, Guang Yang, Pietro Lio
Summary: This paper introduces a novel approach for medical image arbitrary-scale super-resolution, MIASSR, which combines meta-learning and generative adversarial networks (GANs) to achieve super-resolution of medical images at any scale of magnification in (1, 4]. MIASSR shows comparable fidelity performance and the best perceptual quality with the smallest model size compared to state-of-the-art SISR algorithms on single-modal and multi-modal MR brain images. In addition, transfer learning enables MIASSR to handle super-resolution tasks of new medical modalities, with the potential to become a foundational pre-/post-processing step in clinical image analysis tasks.
INTERNATIONAL JOURNAL OF NEURAL SYSTEMS
(2021)
Article
Computer Science, Interdisciplinary Applications
Zhaoyang Song, Xiaoqiang Zhao, Yongyong Hui, Hongmei Jiang
Summary: This study proposes a progressive back-projection network (PBPN) for COVID-CT super-resolution, which achieves better performance in terms of resolution and quality by utilizing back-projection, deep feature extraction, and upscaling in two stages.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Chaofeng Chen, Dihong Gong, Hao Wang, Zhifeng Li, Kwan-Yee K. Wong
Summary: This paper introduces a novel SPatial Attention Residual Network (SPARNet) built on Face Attention Units (FAUs) for face super-resolution, which effectively extracts key features of facial structures by introducing a spatial attention mechanism. Through quantitative comparisons and the introduction of multi-scale discriminators, the method demonstrates superiority in various metrics and the ability to generate high-resolution images.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Interdisciplinary Applications
Bochao Zhao, Nishank Saxena, Ronny Hofmann, Chaitanya Pradhan, Amie Hows
Summary: The current hardware configuration of micro-CT detectors limits the resolution of rock pores that can be achieved. Super resolution techniques and image quality evaluation based on pore throat resolution are proposed to overcome this limitation. Additionally, the use of registered micro-CT images and the ratio of pore throat size to voxel size as grouping criteria in the training dataset are suggested to improve image sharpness and contrast while refining voxel size beyond current technology capabilities.
COMPUTERS & GEOSCIENCES
(2023)
Review
Optics
Khushboo Singla, Rajoo Pandey, Umesh Ghanekar
Summary: This article introduces the single image super resolution (SISR) technique based on generative adversarial networks (GAN), which uses a generator and discriminator network to generate high-resolution images. Different GAN models are classified based on architecture, algorithms, and loss functions, and the research gaps and possible solutions for existing methods are discussed.
Article
Biology
Maira B. H. Moran, Marcelo D. B. Faria, Gilson A. Giraldi, Luciana F. Bastos, Aura Conci
Summary: This study proposes the use of deep learning to improve the quality and resolution of periapical radiographs, with results showing that the SRGAN models with transfer learning outperform other super-resolution methods in terms of Mean Square Error, Peak Signal to Noise Ratio, Structural Similarity Index, and Mean Opinion Score. Visual analysis also confirms the high quality achieved by the SRGAN models, with clearer edge details and fewer blur effects.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Zhineng Chen, Jing Wang, Caiyan Jia, Xiongjun Ye
Summary: This paper proposes a model for super-resolution of pathological images and addresses the challenges of existing studies by constructing a real-image database to validate the effectiveness of the model.
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2023)
Review
Environmental Sciences
Chi Chen, Yongcheng Wang, Ning Zhang, Yuxi Zhang, Zhikang Zhao
Summary: This paper provides a comprehensive review of deep learning-based hyperspectral image super-resolution techniques, including upsampling frameworks, methods, network design, loss functions, representative works, and future directions. The advantages and limitations of 2D and 3D convolutions in HSI SR are analyzed through comparative experiments. Additionally, common datasets, evaluation metrics, and traditional SR algorithms are briefly discussed. To the best of our knowledge, this is the first review on DL-based HSI SR.
Article
Engineering, Electrical & Electronic
Ziyang Liu, Zhengguo Li, Xingming Wu, Zhong Liu, Weihai Chen
Summary: This paper introduces a method that applies generative adversarial network (GAN) to perceptual image super-resolution (SISR). The researchers propose a model-based algorithm that can efficiently extract the detail layer of an image and incorporate it into the GAN to generate more realistic details. Experimental results demonstrate that this method outperforms others in terms of perceptual metrics.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Geochemistry & Geophysics
Sen Jia, Zhihao Wang, Qingquan Li, Xiuping Jia, Meng Xu
Summary: The article introduces a multiattention GAN (MA-GAN) based on generative adversarial networks for generating high-resolution remote sensing images. The network utilizes pyramid convolutional residual dense, attention-based upsampling, and attention-based fusion modules, as well as a loss function to achieve image super-resolution. Experimental results consistently demonstrate the effectiveness of the proposed MA-GAN approach in various remote sensing scenes.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)