Pub Date : 2025-12-15DOI: 10.1109/TCI.2025.3644248
Yodai Suzuki;Ryosuke Isono;Shunsuke Ono
We propose a general deep plug-and-play (PnP) algorithm with a theoretical convergence guarantee. PnP strategies have demonstrated outstanding performance in various image restoration tasks by exploiting the powerful priors underlying Gaussian denoisers. However, existing PnP methods often lack theoretical convergence guarantees under realistic assumptions due to their ad-hoc nature, resulting in inconsistent behavior. Moreover, even when convergence guarantees are provided, they are typically designed for specific settings or require a considerable computational cost in handling non-quadratic data-fidelity terms and additional constraints, which are key components in many image restoration scenarios. To tackle these challenges, we integrate the PnP paradigm with primal-dual splitting (PDS), an efficient proximal splitting methodology for solving a wide range of convex optimization problems, and develop a general convergent PnP framework. Specifically, we establish theoretical conditions for the convergence of the proposed PnP algorithm under a reasonable assumption. Furthermore, we show that the problem solved by the proposed PnP algorithm is not a standard convex optimization problem but a more general monotone inclusion problem, where we provide a mathematical representation of the solution set. Our approach efficiently handles a broad class of image restoration problems with guaranteed theoretical convergence. Numerical experiments on specific image restoration tasks validate the practicality and effectiveness of our theoretical results.
{"title":"Convergent Primal-Dual Plug-and-Play Image Restoration: A General Algorithm and Applications","authors":"Yodai Suzuki;Ryosuke Isono;Shunsuke Ono","doi":"10.1109/TCI.2025.3644248","DOIUrl":"https://doi.org/10.1109/TCI.2025.3644248","url":null,"abstract":"We propose a general deep plug-and-play (PnP) algorithm with a theoretical convergence guarantee. PnP strategies have demonstrated outstanding performance in various image restoration tasks by exploiting the powerful priors underlying Gaussian denoisers. However, existing PnP methods often lack theoretical convergence guarantees under realistic assumptions due to their ad-hoc nature, resulting in inconsistent behavior. Moreover, even when convergence guarantees are provided, they are typically designed for specific settings or require a considerable computational cost in handling non-quadratic data-fidelity terms and additional constraints, which are key components in many image restoration scenarios. To tackle these challenges, we integrate the PnP paradigm with primal-dual splitting (PDS), an efficient proximal splitting methodology for solving a wide range of convex optimization problems, and develop a general convergent PnP framework. Specifically, we establish theoretical conditions for the convergence of the proposed PnP algorithm under a reasonable assumption. Furthermore, we show that the problem solved by the proposed PnP algorithm is not a standard convex optimization problem but a more general monotone inclusion problem, where we provide a mathematical representation of the solution set. Our approach efficiently handles a broad class of image restoration problems with guaranteed theoretical convergence. Numerical experiments on specific image restoration tasks validate the practicality and effectiveness of our theoretical results.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"142-157"},"PeriodicalIF":4.8,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11299501","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830973","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Photoacoustic microscopy (PAM) is a novel implementation of photoacoustic imaging (PAI) for visualizing the 3D bio-structure, which is realized by raster scanning of the tissue. However, temporal resolution, lateral resolution, and penetration depth, as three involved critical imaging parameters, have mutual effect to one the other. The improvement of one parameter results in the degradation of other two parameters, which constrains the overall performance of the PAM system. In this work, we propose to break these limitations by hardware and software co-design. Starting with low lateral resolution, low sampling rate AR-PAM imaging which possesses deep penetration capability, we aim to enhance the lateral resolution and upscale the images, so that high temporal resolution (scanning speed), high lateral resolution, and deep penetration for the PAM system can be achieved. Considering huge information gap between input image and target image, a dedicated dual branch network is proposed, which includes a high resolution branch and a high speed branch to fully extract embedded information from training data. What's more, to effectively fuse the high level semantic information and low level spatial details from these two branches, a novel self-attention and graph induced fusion module is designed. This module significantly eliminates blurring effect to enhance imaging resolution and amplifies the true signal to increase imaging contrast, as proved by comparison algorithms. As a result, the imaging speed is increased 16× times and the imaging lateral resolution is improved 5× times, while the deep penetration merit of AR-PAM modality is still reserved.
{"title":"High Temporal-Lateral Resolution Photoacoustic Microscopy Imaging With Dual Branch Graph Induced Fusion Network","authors":"Zhengyuan Zhang;Xiangjun Yin;Zhuoyi Lin;Wenwen Zhang;Ze Feng;Arunima Sharma;Manojit Pramanik;Chia-Wen Lin;Yuanjin Zheng","doi":"10.1109/TCI.2025.3643726","DOIUrl":"https://doi.org/10.1109/TCI.2025.3643726","url":null,"abstract":"Photoacoustic microscopy (PAM) is a novel implementation of photoacoustic imaging (PAI) for visualizing the 3D bio-structure, which is realized by raster scanning of the tissue. However, temporal resolution, lateral resolution, and penetration depth, as three involved critical imaging parameters, have mutual effect to one the other. The improvement of one parameter results in the degradation of other two parameters, which constrains the overall performance of the PAM system. In this work, we propose to break these limitations by hardware and software co-design. Starting with low lateral resolution, low sampling rate AR-PAM imaging which possesses deep penetration capability, we aim to enhance the lateral resolution and upscale the images, so that high temporal resolution (scanning speed), high lateral resolution, and deep penetration for the PAM system can be achieved. Considering huge information gap between input image and target image, a dedicated dual branch network is proposed, which includes a high resolution branch and a high speed branch to fully extract embedded information from training data. What's more, to effectively fuse the high level semantic information and low level spatial details from these two branches, a novel self-attention and graph induced fusion module is designed. This module significantly eliminates blurring effect to enhance imaging resolution and amplifies the true signal to increase imaging contrast, as proved by comparison algorithms. As a result, the imaging speed is increased 16× times and the imaging lateral resolution is improved 5× times, while the deep penetration merit of AR-PAM modality is still reserved.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"73-85"},"PeriodicalIF":4.8,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830966","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-11DOI: 10.1109/TCI.2025.3643313
Qiushi Huang;Chunzhao Li;Anqi Xiao;Jie Tian;Zhenhua Hu
Fluorescence molecular tomography (FMT), particularly within the second near-infrared window (NIR-II, 1000-1700 nm), is a sophisticated imaging technique for numerous medical applications, enabling reconstruction of the threedimensional distribution of internal tumors from surface fluorescence signals. Recent studies have demonstrated the effectiveness of deep learning methods in FMT reconstruction tasks; however, their performance heavily relies on large-scale, diverse labeled datasets. The existing research primarily focuses on datasets with static tumor characteristics, including fixed tumor numbers, locations, and sizes, which show an insufficient pattern diversity, limiting the neural networks' generalization ability for complex real-world scenarios. To address this limitation, we draw inspiration from the similarity between Monte Carlo photon simulation and the sampling process of diffusion model, to propose a diffusion-based data augmentation strategy. Further, we introduce a novel NIR-II-specific guidance mechanism to enhance sample fidelity and diversity by incorporating spectral optical properties. Quantitative analysis validated that high-quality NIR-II fluorescence signal samples are synthesized, where the proposed NIR-II guidance achieved a 56.7% reduction in Fréchet Inception Distance and a 21.5% improvement in Inception Score, covering a broad spectrum of patterns. Since the synthetic samples are unlabeled, they are integrated with the original dataset to train FMT neural networks using semi-supervised learning. By combining patterndiversifying strengths of diffusion models with semi-supervised learning, the proposed strategy maximizes the utility of limited datasets. Both simulative and in vivo experiments confirmed that data augmentation significantly enhances the network's reconstruction performance in precisely localizing tumor sources and reconstructing complex morphologies.
{"title":"DANG: Data Augmentation Based on NIR-II Guided Diffusion Model for Fluorescence Molecular Tomography","authors":"Qiushi Huang;Chunzhao Li;Anqi Xiao;Jie Tian;Zhenhua Hu","doi":"10.1109/TCI.2025.3643313","DOIUrl":"https://doi.org/10.1109/TCI.2025.3643313","url":null,"abstract":"Fluorescence molecular tomography (FMT), particularly within the second near-infrared window (NIR-II, 1000-1700 nm), is a sophisticated imaging technique for numerous medical applications, enabling reconstruction of the threedimensional distribution of internal tumors from surface fluorescence signals. Recent studies have demonstrated the effectiveness of deep learning methods in FMT reconstruction tasks; however, their performance heavily relies on large-scale, diverse labeled datasets. The existing research primarily focuses on datasets with static tumor characteristics, including fixed tumor numbers, locations, and sizes, which show an insufficient pattern diversity, limiting the neural networks' generalization ability for complex real-world scenarios. To address this limitation, we draw inspiration from the similarity between Monte Carlo photon simulation and the sampling process of diffusion model, to propose a diffusion-based data augmentation strategy. Further, we introduce a novel NIR-II-specific guidance mechanism to enhance sample fidelity and diversity by incorporating spectral optical properties. Quantitative analysis validated that high-quality NIR-II fluorescence signal samples are synthesized, where the proposed NIR-II guidance achieved a 56.7% reduction in Fréchet Inception Distance and a 21.5% improvement in Inception Score, covering a broad spectrum of patterns. Since the synthetic samples are unlabeled, they are integrated with the original dataset to train FMT neural networks using semi-supervised learning. By combining patterndiversifying strengths of diffusion models with semi-supervised learning, the proposed strategy maximizes the utility of limited datasets. Both simulative and in vivo experiments confirmed that data augmentation significantly enhances the network's reconstruction performance in precisely localizing tumor sources and reconstructing complex morphologies.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"128-141"},"PeriodicalIF":4.8,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830964","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Depth estimation plays a great potential role in obstacle avoidance and navigation for further Mars exploration missions. Compared to traditional stereo matching, learning-based stereo depth estimation provides a data-driven approach to infer dense and precise depth maps from stereo image pairs. However, these methods always suffer performance degradation in environments with sparse textures and lacking geometric constraints, such as the unstructured terrain of Mars. To address these challenges, we propose M3Depth, a depth estimation model tailored for Mars rovers. Considering the sparse and smooth texture of Martian terrain, which is primarily composed of low-frequency features, our model incorporates a convolutional kernel based on wavelet transform that effectively captures low-frequency response and expands the receptive field. Additionally, we introduce a consistency loss that explicitly models the complementary relationship between depth map and surface normal map, utilizing the surface normal as a geometric constraint to enhance the accuracy of depth estimation. Besides, a pixel-wise refinement module with mutual boosting mechanism is designed to iteratively refine both depth and surface normal predictions. Experimental results on synthetic Mars datasets with depth annotations show that M3Depth achieves a 16% improvement in depth estimation accuracy compared to other state-of-the-art methods in depth estimation. Furthermore, the model demonstrates strong applicability in real-world Martian scenarios, offering a promising solution for future Mars exploration missions.
{"title":"M3Depth: Wavelet-Enhanced Depth Estimation on Mars via Mutual Boosting of Dual-Modal Data","authors":"Junjie Li;Jiawei Wang;Miyu Li;Yu Liu;Yumei Wang;Haitao Xu","doi":"10.1109/TCI.2025.3642761","DOIUrl":"https://doi.org/10.1109/TCI.2025.3642761","url":null,"abstract":"Depth estimation plays a great potential role in obstacle avoidance and navigation for further Mars exploration missions. Compared to traditional stereo matching, learning-based stereo depth estimation provides a data-driven approach to infer dense and precise depth maps from stereo image pairs. However, these methods always suffer performance degradation in environments with sparse textures and lacking geometric constraints, such as the unstructured terrain of Mars. To address these challenges, we propose M<sup>3</sup>Depth, a depth estimation model tailored for Mars rovers. Considering the sparse and smooth texture of Martian terrain, which is primarily composed of low-frequency features, our model incorporates a convolutional kernel based on wavelet transform that effectively captures low-frequency response and expands the receptive field. Additionally, we introduce a consistency loss that explicitly models the complementary relationship between depth map and surface normal map, utilizing the surface normal as a geometric constraint to enhance the accuracy of depth estimation. Besides, a pixel-wise refinement module with mutual boosting mechanism is designed to iteratively refine both depth and surface normal predictions. Experimental results on synthetic Mars datasets with depth annotations show that M<sup>3</sup>Depth achieves a 16% improvement in depth estimation accuracy compared to other state-of-the-art methods in depth estimation. Furthermore, the model demonstrates strong applicability in real-world Martian scenarios, offering a promising solution for future Mars exploration missions.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"158-171"},"PeriodicalIF":4.8,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hand-held light field (LF) cameras often exhibit low spatial resolution due to the inherent trade-off between spatial and angular dimensions. Existing supervised learning-based LF spatial super-resolution (SR) methods, which rely on pre-defined image degradation models, struggle to overcome the domain gap between the training phase—where LFs with natural resolution are used as ground truth—and the inference phase, which aims to reconstruct higher-resolution LFs, especially when applied to real-world data. To address this challenge, this paper introduces a novel self-supervised learning-based method for LF spatial SR, which can produce higher spatial resolution LF images than originally captured ones without pre-defined image degradation models. The self-supervised method incorporates a hybrid LF imaging prototype, a real-world hybrid LF dataset, and a self-supervised LF spatial SR framework. The prototype makes reference image pairs between low-resolution central-view sub-aperture images and high-resolution (HR) images. The self-supervised framework consists of a well-designed LF spatial SR network with hybrid input, a central-view synthesis network with an HR-aware loss that enables side-view sub-aperture images to learn high-frequency information from the only HR central view reference image, and a backward degradation network with an epipolar-plane image gradient loss to preserve LF parallax structures. Extensive experiments on both simulated and real-world datasets demonstrate the significant superiority of our approach over state-of-the-art ones in reconstructing higher spatial resolution LF images without pre-defined degradation.
{"title":"Self-Supervised Learning-Based Reconstruction of High-Resolution 4D Light Fields","authors":"Jianxin Lei;Dongze Wu;Chengcai Xu;Hongcheng Gu;Guangquan Zhou;Junhui Hou;Ping Zhou","doi":"10.1109/TCI.2025.3642236","DOIUrl":"https://doi.org/10.1109/TCI.2025.3642236","url":null,"abstract":"Hand-held light field (LF) cameras often exhibit low spatial resolution due to the inherent trade-off between spatial and angular dimensions. Existing supervised learning-based LF spatial super-resolution (SR) methods, which rely on pre-defined image degradation models, struggle to overcome the domain gap between the training phase—where LFs with natural resolution are used as ground truth—and the inference phase, which aims to reconstruct higher-resolution LFs, especially when applied to real-world data. To address this challenge, this paper introduces a novel self-supervised learning-based method for LF spatial SR, which can produce higher spatial resolution LF images than originally captured ones without pre-defined image degradation models. The self-supervised method incorporates a hybrid LF imaging prototype, a real-world hybrid LF dataset, and a self-supervised LF spatial SR framework. The prototype makes reference image pairs between low-resolution central-view sub-aperture images and high-resolution (HR) images. The self-supervised framework consists of a well-designed LF spatial SR network with hybrid input, a central-view synthesis network with an HR-aware loss that enables side-view sub-aperture images to learn high-frequency information from the only HR central view reference image, and a backward degradation network with an epipolar-plane image gradient loss to preserve LF parallax structures. Extensive experiments on both simulated and real-world datasets demonstrate the significant superiority of our approach over state-of-the-art ones in reconstructing higher spatial resolution LF images without pre-defined degradation.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"114-127"},"PeriodicalIF":4.8,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-10DOI: 10.1109/TCI.2025.3641749
{"title":"List of Reviewers","authors":"","doi":"10.1109/TCI.2025.3641749","DOIUrl":"https://doi.org/10.1109/TCI.2025.3641749","url":null,"abstract":"","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"11 ","pages":"1682-1685"},"PeriodicalIF":4.8,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11296854","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-08DOI: 10.1109/TCI.2025.3641031
Mengying Jin;Liang Xiao;Zhihui Wei
Hyperspectral image processing faces significant challenges in storage and computation. Snapshot Compressive Imaging (SCI) effectively encodes three-dimensional data into two-dimensional measurements, facilitating efficient data acquisition. However, reconstructing high-quality data from these compressed measurements remains a formidable task. Binary Neural Networks (BNNs) have gained attention for their ability to reduce storage requirements and computational costs. Yet, they often struggle with accuracy loss, fixed quantization limits, and lack of domain knowledge utilization. To overcome these limitations, distribution-adaptive hierarchical quantization-enhanced binary networks are proposed to achieve efficient SCI reconstruction. First, an adaptive distribution strategy and a binary weight evaluation branch are proposed to improve representation accuracy. Second, a hierarchical quantization scheme is presented to enhance multiscale feature extraction while maintaining efficiency. Third, domain-specific priors and a novel sparsity constraint are incorporated to capture fine details and improve training stability. The experimental results demonstrate the superiority of our approach, achieving an increase of 1.98 dB in PSNR and an improvement of 0.055 in SSIM compared to state-of-the-art BNNs.
{"title":"Distribution-Adaptive Hierarchical Quantization Enhanced Binary Networks for Spectral Compressive Imaging","authors":"Mengying Jin;Liang Xiao;Zhihui Wei","doi":"10.1109/TCI.2025.3641031","DOIUrl":"https://doi.org/10.1109/TCI.2025.3641031","url":null,"abstract":"Hyperspectral image processing faces significant challenges in storage and computation. Snapshot Compressive Imaging (SCI) effectively encodes three-dimensional data into two-dimensional measurements, facilitating efficient data acquisition. However, reconstructing high-quality data from these compressed measurements remains a formidable task. Binary Neural Networks (BNNs) have gained attention for their ability to reduce storage requirements and computational costs. Yet, they often struggle with accuracy loss, fixed quantization limits, and lack of domain knowledge utilization. To overcome these limitations, distribution-adaptive hierarchical quantization-enhanced binary networks are proposed to achieve efficient SCI reconstruction. First, an adaptive distribution strategy and a binary weight evaluation branch are proposed to improve representation accuracy. Second, a hierarchical quantization scheme is presented to enhance multiscale feature extraction while maintaining efficiency. Third, domain-specific priors and a novel sparsity constraint are incorporated to capture fine details and improve training stability. The experimental results demonstrate the superiority of our approach, achieving an increase of 1.98 dB in PSNR and an improvement of 0.055 in SSIM compared to state-of-the-art BNNs.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"86-101"},"PeriodicalIF":4.8,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830968","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-05DOI: 10.1109/TCI.2025.3640864
Liliana Borcea;Josselin Garnier
We introduce a method for Multiple Input Multiple Output (MIMO) radar imaging of moving targets in a strongly reflecting, complex stationary scenery (clutter). The radar system has fixed nearby antennas that play the dual role of sources and receivers. It gathers data either by emitting probing pulses from one antenna at a time, or by sending from all the antennas non-coherent, possibly orthogonal, waveforms. We show how to obtain from the measurements an imaging function that depends on search position and velocity and is approximately separable in these variables, for a single moving target. For multiple moving targets in clutter, the imaging function is a sum of separable functions. By sampling this imaging function on a position-velocity grid we obtain an imaging matrix whose Singular Value Decomposition (SVD) allows the separation of the clutter and the targets moving at different velocities. The decomposition also leads directly to estimates of the locations and motion of the targets. The imaging method is designed to work in strong clutter, with unknown and possibly heterogeneous statistics. It does not require prior estimation of the covariance matrix of the clutter response or of its rank. We give an analysis of the imaging method and illustrate how it works with numerical simulations.
{"title":"Moving Targets Imaging by SVD of a Space-Velocity MIMO Radar Data Driven Matrix","authors":"Liliana Borcea;Josselin Garnier","doi":"10.1109/TCI.2025.3640864","DOIUrl":"https://doi.org/10.1109/TCI.2025.3640864","url":null,"abstract":"We introduce a method for Multiple Input Multiple Output (MIMO) radar imaging of moving targets in a strongly reflecting, complex stationary scenery (clutter). The radar system has fixed nearby antennas that play the dual role of sources and receivers. It gathers data either by emitting probing pulses from one antenna at a time, or by sending from all the antennas non-coherent, possibly orthogonal, waveforms. We show how to obtain from the measurements an imaging function that depends on search position and velocity and is approximately separable in these variables, for a single moving target. For multiple moving targets in clutter, the imaging function is a sum of separable functions. By sampling this imaging function on a position-velocity grid we obtain an imaging matrix whose Singular Value Decomposition (SVD) allows the separation of the clutter and the targets moving at different velocities. The decomposition also leads directly to estimates of the locations and motion of the targets. The imaging method is designed to work in strong clutter, with unknown and possibly heterogeneous statistics. It does not require prior estimation of the covariance matrix of the clutter response or of its rank. We give an analysis of the imaging method and illustrate how it works with numerical simulations.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"172-186"},"PeriodicalIF":4.8,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929445","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-24DOI: 10.1109/TCI.2025.3636743
Chunyan Liu;Dianlin Hu;Jiangjun Peng;Hong Wang;Qianyu Shu;Jianjun Wang
Spectral computed tomography (CT) is an imaging technology that utilizes the absorption characteristics of different X-ray energies to obtain the X-ray attenuation characteristics of objects in different energy ranges. However, the limited number of photons detected by spectral CT under a specific X-ray spectrum leads to obvious projection data noise. Making full use of the various properties of the original data is an effective way to recover a clean image from a small amount of noisy projection data. This paper proposes a spectral CT reconstruction method based on representative coefficient image denoising under a low-rank decomposition framework. This method integrates model-driven internal low-rank and nonlocal priors, and data-driven external deep priors, aiming to fully exploit the inherent spectral correlation, nonlocal self-similarity and deep spatial features in spectral CT images. Specifically, we use low-rank decomposition to characterize the global low-rankness of spectral CT images under a plug-and-play framework, and jointly utilize nonlocal low-rankness and smoothness as well as deep image priors to denoise representative coefficient images. Therefore, the proposed method faithfully represents the real underlying information of images by cleverly combining internal and external, nonlocal and local priors. Meanwhile, we design an effective proximal alternating minimization (PAM) algorithm to solve the proposed reconstruction model and establish the theoretical guarantee of the proposed algorithm. Experimental results show that compared with existing popular algorithms, the proposed method can significantly reduce running time while improving spectral CT images quality.
{"title":"Convergence-Guaranteed Spectral CT Reconstruction via Internal and External Prior Mining","authors":"Chunyan Liu;Dianlin Hu;Jiangjun Peng;Hong Wang;Qianyu Shu;Jianjun Wang","doi":"10.1109/TCI.2025.3636743","DOIUrl":"https://doi.org/10.1109/TCI.2025.3636743","url":null,"abstract":"Spectral computed tomography (CT) is an imaging technology that utilizes the absorption characteristics of different X-ray energies to obtain the X-ray attenuation characteristics of objects in different energy ranges. However, the limited number of photons detected by spectral CT under a specific X-ray spectrum leads to obvious projection data noise. Making full use of the various properties of the original data is an effective way to recover a clean image from a small amount of noisy projection data. This paper proposes a spectral CT reconstruction method based on representative coefficient image denoising under a low-rank decomposition framework. This method integrates model-driven internal low-rank and nonlocal priors, and data-driven external deep priors, aiming to fully exploit the inherent spectral correlation, nonlocal self-similarity and deep spatial features in spectral CT images. Specifically, we use low-rank decomposition to characterize the global low-rankness of spectral CT images under a plug-and-play framework, and jointly utilize nonlocal low-rankness and smoothness as well as deep image priors to denoise representative coefficient images. Therefore, the proposed method faithfully represents the real underlying information of images by cleverly combining internal and external, nonlocal and local priors. Meanwhile, we design an effective proximal alternating minimization (PAM) algorithm to solve the proposed reconstruction model and establish the theoretical guarantee of the proposed algorithm. Experimental results show that compared with existing popular algorithms, the proposed method can significantly reduce running time while improving spectral CT images quality.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"46-59"},"PeriodicalIF":4.8,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145766180","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-24DOI: 10.1109/TCI.2025.3636746
Yining Zhang;Jixia Fan;Yanqi Liu;Xinhua Mao
In large-scene synthetic aperture radar (SAR) imaging, the selection of appropriate algorithms is crucial as it directly impacts processing efficiency and image fidelity. The Polar Format Algorithm (PFA) is widely used for its high-speed image formation capabilities. However, its reliance on the planar wavefront approximation inevitably introduces phase errors. A primary challenge arising from linear components of these errors is geometric distortion, which manifests as space-variant shift from actual positions. Traditional inverse warping correction method based on two-dimensional(2-D) interpolation suffers from high computational costs. To address this limitation, this paper proposes a separable 2-D interpolation framework that decouples the correction process into two one-dimensional (1-D) interpolations along azimuth and range axes. Through inverse solutions of geometric distortion functions, it is demonstrated that applying this framework to geometric distortion correction can effectively reduce the complexity while preserving image precision. Simulations and real-data comparisons validate that the proposed fast geometric distortion correction method significantly improve correction speed thus boosting overall computational efficiency.
在大场景合成孔径雷达（SAR）成像中，算法的选择至关重要，它直接影响到处理效率和图像保真度。极坐标格式算法（Polar Format Algorithm， PFA）因其高速图像生成能力而得到广泛应用。然而，它对平面波前近似的依赖不可避免地引入了相位误差。由这些误差的线性分量引起的一个主要挑战是几何畸变，它表现为与实际位置的空间变异偏移。传统的基于二维（2-D）插值的逆翘曲校正方法计算量大。为了解决这一限制，本文提出了一个可分离的二维插值框架，该框架将校正过程解耦为沿方位轴和距离轴的两个一维（一维）插值。通过几何畸变函数的反解，证明将该框架应用于几何畸变校正可以有效地降低校正复杂度，同时保持图像精度。仿真和实际数据对比验证了所提出的快速几何畸变校正方法显著提高了校正速度，从而提高了整体计算效率。
{"title":"Fast Correction for Geometric Distortion in PFA Wavefront Curvature Compensation","authors":"Yining Zhang;Jixia Fan;Yanqi Liu;Xinhua Mao","doi":"10.1109/TCI.2025.3636746","DOIUrl":"https://doi.org/10.1109/TCI.2025.3636746","url":null,"abstract":"In large-scene synthetic aperture radar (SAR) imaging, the selection of appropriate algorithms is crucial as it directly impacts processing efficiency and image fidelity. The Polar Format Algorithm (PFA) is widely used for its high-speed image formation capabilities. However, its reliance on the planar wavefront approximation inevitably introduces phase errors. A primary challenge arising from linear components of these errors is geometric distortion, which manifests as space-variant shift from actual positions. Traditional inverse warping correction method based on two-dimensional(2-D) interpolation suffers from high computational costs. To address this limitation, this paper proposes a separable 2-D interpolation framework that decouples the correction process into two one-dimensional (1-D) interpolations along azimuth and range axes. Through inverse solutions of geometric distortion functions, it is demonstrated that applying this framework to geometric distortion correction can effectively reduce the complexity while preserving image precision. Simulations and real-data comparisons validate that the proposed fast geometric distortion correction method significantly improve correction speed thus boosting overall computational efficiency.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"12 ","pages":"60-72"},"PeriodicalIF":4.8,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145766194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：