Article
Biology
Jianglan Wang, Yong-Jie Li, Kai-Fu Yang
Summary: The paper proposes a simple and efficient method for retinal fundus image enhancement, which divides the input image into three layers and conducts illumination correction, detail enhancement, and denoising at each layer. The method outperforms other related methods and can simultaneously handle tasks of illumination correction, detail enhancement, and noise suppression.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Engineering, Electrical & Electronic
Bo Yang, Dong Pan, Zhaohui Jiang, Jiancai Huang, Weihua Gui
Summary: This paper presents a novel cross-scale decomposition method (CSDM) for low-light image enhancement, which captures small-scale texture and heavy noise and preserves large-scale structure using a Gaussian regularization based on cross-scale relative relationship. Experimental results demonstrate the advantages of this method from both qualitative and quantitative perspectives.
Article
Engineering, Electrical & Electronic
Yun Liu, Anzhi Wang, Hao Zhou, Pengfei Jia
Summary: This paper presents an effective single image dehazing framework based on image decomposition for nighttime hazy images, which out-performs several state-of-the-art dehazing techniques, especially in terms of noise suppression. The proposed algorithm is also capable of handling daytime hazy images and low-light degraded images.
Article
Computer Science, Artificial Intelligence
Yucheng Lu, Seung-Won Jung
Summary: In this paper, a low-light imaging framework that performs joint illumination adjustment, color enhancement, and denoising is proposed to tackle the challenging problem of low-light imaging on mobile devices. The framework avoids the need for massive data recollection, making it more practical and generalizable in real-world settings.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Yu Zhang, Xiaoguang Di, Bin Zhang, Ruihang Ji, Chunhui Wang
Summary: In this paper, a low-light image enhancement method is proposed that combines supervised learning and HSV or Retinex model-based image enhancement methods. A data-driven conditional re-enhancement network is introduced to enhance contrast, brightness, reduce noise and color distortion in low-light images. Experimental results demonstrate the effectiveness of the proposed method and its superiority when compared to other image enhancement methods.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Xiaosong Li, Fuqiang Zhou, Haishu Tan
Summary: The proposed image fusion method is based on three-layer decomposition and sparse representation, which effectively fuses and denoises high-frequency components by adaptively designing sparse reconstruct error parameters. Additionally, the structure-texture decomposition model and carefully designed fusion rules are used to fully utilize details and energy in low-frequency components. The experimental results show that this method can effectively address clean and noisy image fusion problems and outperform some state-of-the-art methods in subjective visual and quantitative evaluations.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Mathematics, Applied
Hossein Khodabakhshi Rafsanjani, Hossein Noori, Nasibe Naseri
Summary: In this paper, the ENI operator, which can distinguish well between impulse noise and signal, is analyzed and a efficient diffusion based method for impulse denoising is proposed. The experimental results confirm the efficiency of the proposed method according to subjective and objective criteria.
COMPUTERS & MATHEMATICS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Jielin Jiang, Xiangming Hong, Yingnan Zhao, Xiaonglong Xu, Yan Cui
Summary: Convolutional neural networks (CNNs) have shown impressive performance in removing additive white Gaussian noise, but are unsatisfactory in removing mixed noise due to limited receptive field. Recent approaches using attention mechanism (AM) to capture global information still lack computational efficiency. This paper proposes a novel model called simple dual attention mechanism UNet (SDAUNet) for mixed noise removal, which uses UNet architecture to acquire multi-scale features and a simple dual attention convolutional block to capture global features efficiently. Experimental results demonstrate the superiority of SDAUNet in measurement metrics and visual performance compared to other state-of-the-art methods.
IET IMAGE PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Liming Tang, Liang Wu, Zhuang Fang, Chunyan Li
Summary: This study introduces a non-convex ternary variational decomposition model that decomposes images into structure, texture, and noise components. It effectively removes noise while preserving image edges and textures, and an efficient solving algorithm is utilized. Numerical results on synthetic and real images validate the model's effectiveness and performance is superior to state-of-the-art models based on PSNR and SSIM metrics.
IET SIGNAL PROCESSING
(2022)
Article
Engineering, Electrical & Electronic
Weidong Zhang, Songlin Jin, Peixian Zhuang, Zheng Liang, Chongyi Li
Summary: In this letter, a method is proposed to address the degradation issues of underwater captured images. It uses piecewise color correction and dual prior optimized contrast enhancement. The method corrects the color cast of each color channel and enhances the contrast and texture detail of underwater images by decomposing the base layer and detail layer in HSV color space. Experimental results show that the method outperforms eleven state-of-the-art methods and has good generalization capability for fog and low-light images.
IEEE SIGNAL PROCESSING LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Yan Cao, Jianchong Wei, Sifan Chen, Baihe Chen, Zhensheng Wang, Zhaohui Liu, Chengbin Chen
Summary: In this article, we propose a joint blind denoising and dehazing method for the recovery and enhancement of remote sensing images. We introduce an efficient noise level estimation method and integrate it into a fast and flexible denoising convolutional neural network for blind denoising. We also utilize a multiscale guided filter to enhance the detailed layers and apply dehazing with a corrected boundary constraint.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Xianglang Wang, Youfang Lin, Shuo Zhang
Summary: We propose a multiple stream progressive restoration network to enhance and denoise light fields (LFs) under low light conditions. Three types of input are designed to fully utilize the supplementary information and preserve the epipolar information. A multi-stream interaction module is developed to aggregate features from different restoration streams. Multiple stages restoration is introduced to gradually reconstruct the LF.
IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING
(2023)
Article
Computer Science, Information Systems
Nguyen Ngoc Hien, Dang Ngoc Hoang Thanh, Ugur Erkan, Joao Manuel R. S. Tavares
Summary: In this article, a salt and pepper noise removal method based on thresholding and regularization techniques is proposed. The experimental results demonstrate its superiority over other methods.
Article
Geochemistry & Geophysics
Mehwish Iqbal, Muhammad Mohsin Riaz, Syed Sohaib Ali, Abdul Ghafoor, Attiq Ahmad
Summary: This letter presents a method of underwater image enhancement using Laplace decomposition. The method decomposes the image into low-frequency and high-frequency bands, and processes them separately to achieve image enhancement.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Engineering, Multidisciplinary
Onur Karaoglu, Hasan Sakir Bilge, Ihsan Uluer
Summary: This study applies deep learning networks to denoise speckle noises in ultrasound images and compares their performance with classical image enhancement algorithms. The results show that the proposed deep learning networks outperform other networks and algorithms in terms of peak signal-to-noise ratio and structural similarity index.
ENGINEERING SCIENCE AND TECHNOLOGY-AN INTERNATIONAL JOURNAL-JESTECH
(2022)
Article
Computer Science, Artificial Intelligence
Keunsoo Ko, Yeong Jun Koh, Chang-Su Kim
Summary: A lightweight blind image denoiser, BCDNet, is proposed in this paper to achieve excellent trade-offs between performance and network complexity. Through the compact denoising network CDNet and the guidance network GNet, BCDNet effectively removes noise adaptively according to the severity while reducing the number of parameters.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Information Systems
Jae-Han Lee, Chang-Su Kim
Summary: This paper proposes a novel algorithm for monocular depth estimation using relative depths, which goes through multiple stages to generate the final depth map, and demonstrates state-of-the-art performance on the NYUv2 dataset.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2022)
Article
Computer Science, Information Systems
Thuong Van Nguyen, An Gia Vien, Chul Lee
Summary: We proposed a real-time dehazing algorithm based on multiscale guided filtering. By constructing an image pyramid and iteratively upsampling, we achieved efficient estimation of transmission map and atmospheric light. Extending the algorithm to real-time video dehazing also reduced flickering artifacts and showed comparable or better performance than state-of-the-art algorithms.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Computer Science, Information Systems
Whan Choi, Yeong Jun Koh, Chang-Su Kim
Summary: This paper proposes a video frame interpolation network based on both symmetric and asymmetric motion-based warping modules, which can effectively handle linear and non-linear motions as well as occlusions. The combination of symmetric and asymmetric warping results improves the reliability of intermediate frame reconstruction.
Proceedings Paper
Computer Science, Artificial Intelligence
An Gia Vien, Chul Lee
Summary: We propose a novel single-shot high dynamic range (HDR) imaging algorithm based on exposure-aware dynamic weighted learning, which outperforms conventional algorithms by using local dynamic filters and exposure-aware feature fusion to reconstruct HDR images.
COMPUTER VISION, ECCV 2022, PT VII
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Seon-Ho Lee, Chang-Su Kim
Summary: The proposed chainization algorithm effectively learns orderings from partially ordered data. By using a binary comparator and extending the Kahn's algorithm, it predicts missing ordering relations and forms a chain representing a linear ordering of instances. Experimental results show excellent performances in various weak supervision scenarios.
COMPUTER VISION, ECCV 2022, PT XIII
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Jinyoung Jun, Jae-Han Lee, Chul Lee, Chang-Su Kim
Summary: This algorithm proposes a method for monocular depth estimation by decomposing a metric depth map into a normalized depth map and scale features. The network consists of a shared encoder and three decoders, which estimate gradient maps, a normalized depth map, and a metric depth map. The algorithm achieves accurate metric depth estimation by using relative depth features and can improve performance even without metric depth labels in the dataset.
COMPUTER VISION - ECCV 2022, PT II
(2022)
Article
Computer Science, Information Systems
Seonghyun Park, Chul Lee
Summary: In this work, a multiscale progressive fusion algorithm (MPFusion) is proposed to effectively extract and fuse multiscale features of infrared and visible images, preserving complementary information while avoiding bias towards either of the source images. The algorithm consists of two networks, IRNet and FusionNet, which extract the intrinsic features of the infrared and visible images, respectively. The multiscale information of the infrared image is transferred from IRNet to FusionNet to generate an informative fusion result. The proposed algorithm utilizes multi-dilated residual blocks and progressive fusion blocks to effectively and adaptively fuse complementary features. Edge-guided attention maps are also employed to preserve complementary edge information during fusion.
Article
Computer Science, Information Systems
Le Thi Hue Dao, Truong Thanh Nhat Mai, Wook Hong, Sanghyun Park, Hokwon Kim, Joon Goo Lee, Min-Seok Kim, Chul Lee
Summary: We propose an orientation prediction algorithm based on Kalman-like error compensation for VR and AR devices, which improves the prediction accuracy by taking into account the accuracies of previous predictions using IMU measurements.
Article
Computer Science, Information Systems
Keunsoo Ko, Chang-Su Kim
Summary: This study proposes an edge-aware interactive contrast enhancement algorithm that allows users to easily adjust image contrast according to their preferences. The algorithm generates an edge-aware mask based on user-provided scribbles and restores an enhanced image through a trained neural network. Experiments show that the proposed algorithm satisfactorily enhances images with simple scribbles and can also produce enhanced images automatically.
Article
Computer Science, Artificial Intelligence
Truong Thanh Nhat Mai, Edmund Y. Lam, Chul Lee
Summary: In this paper, an algorithm unrolling approach to ghost-free HDR image synthesis algorithm is proposed. It unrolls an iterative low-rank tensor completion algorithm into deep neural networks to combine the advantages of both learning- and model-based approaches while overcoming their weaknesses. Experimental results show that the proposed algorithm outperforms state-of-the-art algorithms in terms of HDR image synthesis performance and robustness, using significantly fewer training samples.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Information Systems
Yuk Heo, Yeong Jun Koh, Chang-Su Kim
Summary: In video object segmentation, the memory-based approach has significantly improved the results. However, the temporal smoothness between the query frame and the local memory is often overlooked. This paper proposes local memory read-and-compare operations to address this issue, leading to more strict segmentation results compared to state-of-the-art algorithms.
Article
Computer Science, Information Systems
Hyunkook Park, An Gia Vien, Hanul Kim, Yeong Jun Koh, Chul Lee
Summary: In this paper, we propose an end-to-end unpaired learning approach for screenshot image demoireing based on cyclic moire learning. The proposed algorithm consists of a moireing network and a demoireing network. Experimental results demonstrate that the proposed algorithm significantly outperforms state-of-the-art unsupervised demoireing algorithms as well as image restoration algorithms.
Article
Computer Science, Information Systems
Whan Choi, Yeong Jun Koh, Chang-Su Kim
Summary: This work proposes a novel video interpolation network by developing a multi-scale warping module to robustly interpolate intermediate frames, which outperforms state-of-the-art video interpolation algorithms on various benchmark datasets.