Article
Computer Science, Artificial Intelligence
Yuanyuan Liu, Mengtao Yue, Han Yan, Lu Zhu
Summary: This study introduces a lightweight Transformer structure into convolutional neural networks for single-image super-resolution problems. The proposed model extracts both internal and external information of the image by using a dense block structure and residual connection to build a residual dense convolution block (RDCB). The lightweight transformer block (LTB) further extracts features and learns texture details between patches through the self-attention mechanism. The effectiveness of the proposed model is demonstrated through extensive evaluations on benchmark datasets.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Jifeng Guo, Feicai Lv, Jiayou Shen, Jing Liu, Mingzhi Wang
Summary: This paper proposes an improved generative adversarial network to enhance the super-resolution reconstruction effect of medium- and low-resolution remote sensing images. By improving the network structure and loss function design, and effectively combining image information from three scales, this method achieves more realistic and reliable texture reconstruction.
IET IMAGE PROCESSING
(2023)
Article
Engineering, Biomedical
Kamlesh Pawar, Gary F. Egan, Zhaolin Chen
Summary: A deep learning method incorporating domain knowledge of parallel magnetic resonance imaging is proposed for accelerated image reconstruction, achieving significant improvements in accuracy and robustness compared to state-of-the-art methods. The method utilizes a novel loss function and outperforms others in various contrasts, showing stability and robustness to perturbations. Comprehensive validation on large datasets demonstrates accurate and stable image enhancement through the regularization of deep learning-based reconstruction with domain knowledge.
COMPUTERIZED MEDICAL IMAGING AND GRAPHICS
(2021)
Article
Computer Science, Artificial Intelligence
Liyuan Ma, Tingwei Gao, Haibin Shen, Kejie Huang
Summary: Person image generation aims to preserve the original human appearance in different poses. The alignment of appearance and pose domains is crucial for achieving this task. Previous alignment methods encounter challenges in producing fine texture details and accurately estimating appearance flows. In this article, the importance of multi-scale alignment in both low-level and high-level domains is demonstrated, and a novel method called Multi-scale Cross-domain Alignment (MCA) is proposed. MCA achieves superior performance on popular datasets, validating the effectiveness of the approach.
CAAI TRANSACTIONS ON INTELLIGENCE TECHNOLOGY
(2023)
Article
Developmental Biology
Dongsun Shin, Mitsutoshi Nakamura, Yoshitaka Morishita, Mototsugu Eiraku, Tomoko Yamakawa, Takeshi Sasamura, Masakazu Akiyama, Mikiko Inaki, Kenji Matsuno
Summary: Proper positioning of nuclei in the Drosophila anterior midgut plays a crucial role in subsequent LR-asymmetric development of the organ. Wnt4 signaling, myosin II, and the LING complex are important for collective nuclear behavior and proper positioning.
Article
Computer Science, Artificial Intelligence
Xiaolin Kong, Tao Gao, Ting Chen, Jing Zhang
Summary: Rain removal in computer vision is a challenging problem, especially for single-image deraining. This paper proposes a progressive dilation dense residual fusion network to remove rain streaks more thoroughly and retain more details. The network is designed in a cascade manner with multiple fusion blocks, and a detail compensation memory mechanism is leveraged to retain more background details. Experimental results demonstrate that the proposed method outperforms previous state-of-the-art methods in rain streaks removal, background details preservation, and image noise removal.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Jianfang Cao, Xiaohui Hu, Hongyan Cui, Yunchuan Liang, Zeyu Chen
Summary: This study proposes a super-resolution reconstruction method for fuzzy murals based on a generative adversarial network with self-attention. Experiments show that the proposed method achieves higher peak signal-to-noise ratio and structural similarity compared to other SR reconstruction algorithms. Subjective evaluation also indicates that the method can better reconstruct the texture details of murals, meeting the visual perception needs of the public.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Dan Xu, Xiaopeng Fan, Debin Zhao, Wen Gao
Summary: Depth map super resolution is improved by a novel multiscale and multidirection framework combined with semantic inference. The framework captures and assembles intrinsic geometrical structures through a multiview non-subsampled contourlet transform, isolating the discontinuities of contours while retaining smoothness along contours. Semantic inference segments and labels the depth map, and a label refinement strategy corrects inaccurate labels. Experimental results show significant improvements over state-of-the-art algorithms.
IET IMAGE PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Guangping Li, Huanling Xiao, Dingkai Liang
Summary: In this paper, an enhanced dual branches network (EDBNet) is proposed for generating arbitrary-scale super-resolution (SR) images. A scale-guidance upsampling module (SGU) is designed to guide the convolution weights by adding scale factors and pixel-level features.
ELECTRONICS LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Bin Kong, Jing Qian, Pinhao Song, Jing Yang, Amir Hussain
Summary: Underwater images often suffer from biased colours and reduced contrast due to light propagation effects in water. Traditional restoration or enhancement methods are time-consuming and often fail to produce satisfactory results. In this paper, we propose a new method that imitates the colour constancy mechanism of human vision using double-opponency to address these issues. Experimental results demonstrate that our method achieves higher quality clarified underwater images with significantly less computational cost compared to existing methods.
CAAI TRANSACTIONS ON INTELLIGENCE TECHNOLOGY
(2023)
Article
Construction & Building Technology
Cheng Liu, Guihua Liu
Summary: This paper analyzes the pore structure parameters of foam concrete using 3D reconstruction technology, combined with image processing techniques, to more accurately quantify the pore structure parameters of foam concrete. This method is expected to be a new means for the analysis of porous materials.
CONSTRUCTION AND BUILDING MATERIALS
(2021)
Article
Optics
Min-kyung Kim, Yoon-soo Yeo, Hyun-joon Shin
Summary: This study simplified the processing of high-resolution fiberscope images by introducing a combination of binarization and local thresholding, reducing the complexity of core peak detection and effectively decreasing the overall processing time.
OPTICS COMMUNICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Dangguo Shao, Li Qin, Yan Xiang, Lei Ma, Hui Xu
Summary: The paper proposes a medical image blind super-resolution model (Med-BSR) based on an improved degradation process to address the high requirements for clinical diagnosis on the resolution of medical images. The model makes the degradation factors in medical image blind SR, such as blur, noise, and downsampling, more complex and practical. The authors use a random select/combine strategy to significantly expand the degradation space by randomly arranging and combining the type and order of each degradation factor. The extensive experimental results demonstrate that the designed model can accurately restore the natural degradation process and reconstruct high-quality SR medical images, while also having good generalization ability to realistic images.
IET IMAGE PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Simon Grosche, Andy Regensky, Alexander Sinn, Jurgen Seiler, Andre Kaup
Summary: Non-regular three-quarter sampling using L-shaped pixels has been shown to improve image sensor quality, and a faster version of the reconstruction algorithm, RL-JSDE, provides significant speedups on both CPU and GPU without sacrificing image quality.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Computer Science, Artificial Intelligence
Thiago da Silva Teixeira, Mauren Louise Sguario Coelho de Andrade, Mathias Rodrigues da Luz
Summary: This study introduces a new method for solving puzzles based on sequence of layers, using the Middle Triangle Method to select corresponding keypoints between puzzle pieces and reference image, leading to a high ratio of correctly positioned fragments reaching up to 100%, surpassing other puzzle solvers.
PATTERN RECOGNITION LETTERS
(2021)
Article
Optics
Igor Shevkunov, Vladimir Katkovnik, Daniel Claus, Giancarlo Pedrini, Nikolay Petrov, Karen Egiazarian
OPTICS AND LASERS IN ENGINEERING
(2020)
Article
Optics
Igor Shevkunov, Vladimir Katkovnik, Karen Egiazarian
Article
Engineering, Electrical & Electronic
M. S. Kulya, V. Ya Katkovnik, K. Egiazarian, N. Petrov
QUANTUM ELECTRONICS
(2020)
Article
Environmental Sciences
Oleg Ieremeiev, Vladimir Lukin, Krzysztof Okarma, Karen Egiazarian
Article
Optics
Seyyed R. M. Rostami, Vladimir Katkovnik, Karen Egiazarian
Summary: An optimal optical transfer function (OTF) is proposed for RGB inverse imaging to achieve extended depth of field and reduced color aberrations. This new inverse imaging technique, demonstrated in optical setups with lens and lensless, shows better performance in a lensless system designed for the wavelength range of 400 to 700 nm and depth of field range from 0.5 to 1000 m.
OPTICAL ENGINEERING
(2021)
Article
Environmental Sciences
Vladimir Lukin, Irina Vasilyeva, Sergey Krivenko, Fangfang Li, Sergey Abramov, Oleksii Rubel, Benoit Vozel, Kacem Chehdi, Karen Egiazarian
Article
Chemistry, Multidisciplinary
Sergey Krivenko, Vladimir Lukin, Olha Krylova, Liudmyla Kryvenko, Karen Egiazarian
Summary: This paper proposes a noniterative approach to the visually lossless compression problem of dental images, focusing on lossy compression, preservation of diagnostic information, and noise characteristics of dental images. By analyzing the dependencies of quality metrics on quantization step and considering distortion visibility thresholds, the compression of images is controlled effectively.
APPLIED SCIENCES-BASEL
(2021)
Article
Optics
Vladimir Katkovnik, Igor Shevkunov, Karen Egiazarian
Summary: The study introduces a phase retrieval algorithm for hyperspectral imaging that avoids random phase coding commonly used in traditional methods. Through shearography optical setup, the algorithm demonstrates excellent performance in object phase and thickness imaging in simulation and experimental tests.
OPTICAL ENGINEERING
(2021)
Article
Optics
Veronica Cazac, Elena Achimova, Vladimir Abashkin, Alexandr Prisacar, Constantin Loshmanschii, Alexei Meshalkin, Karen Egiazarian
Summary: This paper presents the fabrication of optical vortex DOEs on photosensitive thin films using analog and digital approaches, and compares the efficiency evolution of the three types of DOEs. The study provides new evidence on the influence of analog and digital generation of spiral wavefront on the performance of vortex DOEs.
Article
Environmental Sciences
Oleksii Rubel, Vladimir Lukin, Andrii Rubel, Karen Egiazarian
Summary: Radar imaging has many advantages, but SAR images are affected by noise-like speckle. The local statistic Lee filter is a popular despeckling technique, and this study demonstrates how filter parameters can be selected based on efficiency prediction using a trained neural network. Adaptive selection of filter window size can significantly improve filtering efficiency.
Article
Optics
Seyyed Reza Miri Rostami, Samuel Pinilla, Igor Shevkunov, Vladimir Katkovnik, Karen Egiazarian
Summary: This paper introduces a power-balanced hybrid optical imaging system with a diffractive computational camera utilizing a refractive lens and multilevel phase mask (MPM). By optimizing the optical power balance and MPM design, the system shows significant advantages in terms of image quality reconstruction and depth-of-field range.
Article
Optics
Peter Kocsis, Igor Shevkunov, Vladimir Katkovnik, Heikki Rekola, Karen Egiazarian
Summary: The proposed method introduces a lensless single-shot phase retrieval approach that separates carrying and object wavefronts. By calibrating discrepancies between computational models and physical elements and implementing pixel super-resolution processing, it reconstructs the object wavefront with correction from the carrying wavefront, demonstrating robustness in simulations and experiments. In phase bio-imaging, it achieves high-quality imaging results and records dynamic scenes efficiently with the single-shot advantage.
Proceedings Paper
Optics
Seyyed Reza Miri Rostami, Samuel Pinilla, Igor Shevkunov, Vladimir Katkovnik, Karen Eguiazarian
Summary: This paper presents a hybrid imaging system that achieves optimized imaging with extended depth of field and reduced chromatic aberrations. A fully differentiable image formation model and neural network techniques are used for imaging optimization. By comparing the effects of different optical parameters on imaging accuracy and quality, the study proposes an application of hybrid optics in compact cameras.
UNCONVENTIONAL OPTICAL IMAGING III
(2022)
Proceedings Paper
Acoustics
Pham Huu Thanh Binh, Cristovao Cruz, Karen Egiazarian
Summary: This paper introduces a learning-based denoising method, FlashLight CNN, which utilizes a deep neural network for image denoising. By combining deep residual networks and inception networks, the proposed approach outperforms current state of the art image denoising methods in quantitative and visual comparisons.
28TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2020)
(2021)
Proceedings Paper
Imaging Science & Photographic Technology
Vladimir Katkovnik, Igor Shevkunov, Karen Egiazarian
2020 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)
(2020)