Article
Engineering, Electrical & Electronic
Yasmin Sarcheshmehpour, Yu Tian, Linli Zhang, Alexander Jung
Summary: This paper focuses on optimization methods for training local models in decentralized datasets with a network structure. By formulating federated learning as generalized total variation (GTV) minimization, the authors propose a flexible approach that extends existing methods. They also introduce a fully decentralized federated learning algorithm and provide an upper bound on the deviation of local model parameters learnt by their algorithm compared to an oracle-based method.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2023)
Article
Engineering, Multidisciplinary
Cheng Kai, Jiang Min, Zhiping Qu, Jianqiao Yu, Sun Yi
Summary: The proposed nonlocal 3D Shearlet sparse regularization method effectively restores structural details of noisy sparse-view CT images, suppresses noise, and protects fine structures, improving the quality of the reconstructed image.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2021)
Article
Computer Science, Software Engineering
Chunxue Wang, Huayan Zhang, Ligang Liu
Summary: This paper introduces a reflectance and illumination decomposition model for the Retinex problem using total generalized variation regularization and H1 decomposition. Experimental results show that the proposed model is comparable both quantitatively and qualitatively with several state-of-the-art methods.
Article
Computer Science, Software Engineering
Chunxue Wang, Linlin Xu, Ligang Liu
Summary: This paper proposes a novel image decomposition model for various image processing applications such as compression, enhancement, and texture removal. The model utilizes non-convex regularization and convolutional sparse coding to accurately decompose images into structure and texture components. It also incorporates structure-aware and texture-aware measures to effectively distinguish between these components.
Article
Mathematics, Applied
Jinlan Li, Zhaoyang Xie, Guoqi Liu, Liu Yang, Jian Zou
Summary: In this paper, a convex-nonconvex graph total variation (CNC-GTV) regularization method is proposed for diffuse optical tomography (DOT) reconstruction. By combining the powerful representation ability of graph and the edge-preserving ability of total variation (TV) regularization, this method solves the issue of underestimating large edge values that classical TV regularization tends to have. The global convexity of the objective function is guaranteed by adjusting the nonconvex control parameters, and an alternating direction multiplier method (ADMM) is used to solve the proposed DOT reconstruction model. Numerical experiments demonstrate the superior performance of the proposed model in terms of visual and numerical results.
MATHEMATICAL METHODS IN THE APPLIED SCIENCES
(2023)
Article
Geochemistry & Geophysics
Zhongqiu Xu, Bingchen Zhang, Zhe Zhang, Mingzhi Wang, Yirong Wu
Summary: This letter proposes a nonconvex-nonlocal TV (NLTV) regularization for improving the reconstruction accuracy of SAR imaging and avoiding over-smoothing of isolated point targets while simultaneously enhancing the features of point and distributed targets. The performance of the proposed method is verified using simulated data and real data.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Mathematics, Applied
Yiming Gao, Zhengmeng Jin, Xu Li
Summary: In this paper, a variational model based on dynamic optimal transportation and total generalized variation is proposed for CT reconstruction. It aims to reduce the radiation dose for patients and improve image quality and structure preservation. The final state image of the optimal transport problem is reconstructed through CT inversion, utilizing the given initial state as a template for structural information. The proposed model is solved numerically using a first-order algorithm based on the primal-dual method and demonstrated to achieve high-quality and structurally preserved image reconstruction in sparse-view CT.
Article
Chemistry, Multidisciplinary
Zhihong Liu, Qingyu Liu, Zunmin Liu, Chao Li, Qixin Xu
Summary: In this paper, a novel 2D-DOA estimation method with total variation regularization is proposed to solve the problem of sparse DOA estimation for spatially extended sources. The proposed method builds an extended sources acoustic model, a two-dimensional array manifold, and utilizes total variation regularization to group the non-zero coefficients together with optimum sparseness. The method shows better robustness to noise, sparsity, and estimation speed with higher resolution compared to traditional methods.
APPLIED SCIENCES-BASEL
(2023)
Article
Engineering, Biomedical
Kaichao Liang, Li Zhang, Yuxiang Xing
Summary: In this study, a new high spatial resolution X-ray diffraction (XRDT) method combining coded-aperture compressed-sensing acquisition and sparse-view scan was proposed. The proposed RotationCA-XRDT method achieved significantly better image resolution than current SnapshotCA-XRDT methods in the field and reduced the scan time from hours to minutes. It has great potential for biological sample XRDT inspection.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Mathematics, Applied
Monica Pragliola, Luca Calatroni, Alessandro Lanza, Fiorella Sgallari
Summary: This research reviews and connects the major contributions in the field of space-variant TV-type image reconstruction models, with a focus on their Bayesian interpretation, paving the way for new and unexplored research directions.
Article
Chemistry, Analytical
Manasavee Lohvithee, Wenjuan Sun, Stephane Chretien, Manuchehr Soleimani
Summary: This paper proposed a computer-aided training method for hyperparameter selection in limited data X-ray computed tomography (XCT) reconstruction using ant colony optimization. The experiments showed that the proposed method outperformed other algorithms in terms of reconstruction quality.
Article
Mathematics, Applied
Jin-Ju Wang, Ting-Zhu Huang, Jie Huang, Liang-Jian Deng
Summary: Sparse hyperspectral unmixing is a hot topic in remote sensing, aiming to find an optimal spectral subset to model mixed pixels in hyperspectral images. The SUnSAL-TV method, incorporating a TV regularizer, shows promising unmixing performance and is solved using the ADMM framework. A new weighted collaborative sparse unmixing model (WCSU-TV) and a two-step iterative strategy based on ADMM are proposed in this paper, demonstrating effectiveness in experiments on simulated and real hyperspectral data.
APPLIED MATHEMATICS AND COMPUTATION
(2021)
Article
Automation & Control Systems
Zhi Wang, Qiang Lin, Yingyi Chen, Ping Zhong
Summary: This study proposes a new block-based multi-view classification model that addresses the issues of ignoring the individuality and relationship of views in existing methods, and uses slack labels to increase the distance between distinct classes.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Mathematics
Kuan Li, Chun Huang, Ziyang Yuan
Summary: This paper investigates error estimations for total variation regularization, which is applied in various fields such as signal processing, imaging, and machine learning. The study focuses on properties of the minimizer for the TV regularization problem, including stability, consistency, and convergence rate. Both a priori and a posteriori rules are considered, with an improved convergence rate based on sparsity assumption. Additionally, the paper discusses non-sparsity conditions commonly found in practice, presenting corresponding convergence rates under mild conditions.
Article
Instruments & Instrumentation
Junaid Ahmed, Guiyun Tian, Abdul Baseer Buriro, Gulsher Baloch, Muhammad Waqas Soomro
Summary: In this paper, a tensor nuclear norm-based low-rank and sparse total variation regularization method is proposed to address the issue of uneven illumination and high-frequency thermal noise in OPT-based inspection. The proposed algorithm removes noise, segments/extracts defect information from thermal video sequences, and improves resolution and contrast.
INFRARED PHYSICS & TECHNOLOGY
(2022)