Pub Date : 2025-12-26DOI: 10.1109/LSP.2025.3648640
Zhen Gao;Yiping Jiang;Rong Yang;Xingqun Zhan
Recovering signals from noisy observations remains challenging due to the ill-posedness of inverse problems. While non-convex regularization methods like the standard Cauchy penalty improve estimation accuracy, it lacks adaptability across diverse scenarios. In response, this letter proposes a fractional-order Cauchy (q-Cauchy) penalty inspired by the Lq maximum likelihood estimation. By introducing the parameter $q$, the q-Cauchy penalty achieves greater adaptability in diverse scenarios. Specifically, we also derive sufficient convexity conditions for its proximal operator and propose a forward-backward solver. Simulation results demonstrate that the q-Cauchy with the appropriate $q$ outperforms the baseline methods in both 1D signal denoising and 2D image deblurring tasks.
{"title":"A Fractional-Order Cauchy Penalty With Enhanced Adaptability for Signal Recovery","authors":"Zhen Gao;Yiping Jiang;Rong Yang;Xingqun Zhan","doi":"10.1109/LSP.2025.3648640","DOIUrl":"https://doi.org/10.1109/LSP.2025.3648640","url":null,"abstract":"Recovering signals from noisy observations remains challenging due to the ill-posedness of inverse problems. While non-convex regularization methods like the standard Cauchy penalty improve estimation accuracy, it lacks adaptability across diverse scenarios. In response, this letter proposes a fractional-order Cauchy (q-Cauchy) penalty inspired by the Lq maximum likelihood estimation. By introducing the parameter <inline-formula><tex-math>$q$</tex-math></inline-formula>, the q-Cauchy penalty achieves greater adaptability in diverse scenarios. Specifically, we also derive sufficient convexity conditions for its proximal operator and propose a forward-backward solver. Simulation results demonstrate that the q-Cauchy with the appropriate <inline-formula><tex-math>$q$</tex-math></inline-formula> outperforms the baseline methods in both 1D signal denoising and 2D image deblurring tasks.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"456-460"},"PeriodicalIF":3.9,"publicationDate":"2025-12-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929583","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-25DOI: 10.1109/LSP.2025.3648249
Menghui Lei;Xiangyang Zeng;Mingmin Zeng;Anqi Jin
The attention mechanism improves underwater acoustic target recognition (UATR) by suppressing irrelevant features. However, due to the uncertainty and scarcity of underwater acoustic target (UWAT) signals, complicated deterministic attention modules increase the risk of model overfitting, resulting in limited improvement or even degradation in the performance of UATR. This letter proposes a Bayesian Hybrid Attention Module (BHAM) that enhances UATR based on time–frequency (T–F) features. BHAM models attention weights as random variables following Beta and Dirichlet distributions to capture uncertainty of UWAT signals and mitigate overfitting, while strengthening T–F feature representation via Bayesian channel attention and Bayesian T–F attention. By learning attention distributions in a Bayesian manner, BHAM effectively models complex dependencies in UWAT signals. Experiments on the DeepShip dataset demonstrate that BHAM alleviates overfitting and generalizes well across different network backbones.
{"title":"A Bayesian Hybrid Attention Module for Underwater Acoustic Target Recognition","authors":"Menghui Lei;Xiangyang Zeng;Mingmin Zeng;Anqi Jin","doi":"10.1109/LSP.2025.3648249","DOIUrl":"https://doi.org/10.1109/LSP.2025.3648249","url":null,"abstract":"The attention mechanism improves underwater acoustic target recognition (UATR) by suppressing irrelevant features. However, due to the uncertainty and scarcity of underwater acoustic target (UWAT) signals, complicated deterministic attention modules increase the risk of model overfitting, resulting in limited improvement or even degradation in the performance of UATR. This letter proposes a Bayesian Hybrid Attention Module (BHAM) that enhances UATR based on time–frequency (T–F) features. BHAM models attention weights as random variables following Beta and Dirichlet distributions to capture uncertainty of UWAT signals and mitigate overfitting, while strengthening T–F feature representation via Bayesian channel attention and Bayesian T–F attention. By learning attention distributions in a Bayesian manner, BHAM effectively models complex dependencies in UWAT signals. Experiments on the DeepShip dataset demonstrate that BHAM alleviates overfitting and generalizes well across different network backbones.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"441-445"},"PeriodicalIF":3.9,"publicationDate":"2025-12-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929434","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-25DOI: 10.1109/LSP.2025.3648638
Fangqing Xiao;Dirk Slock
The Cramér–Rao bound (CRB) quantifies the variance lower bound for unbiased estimators, but it is intractable to evaluate in linear hierarchical Bayesian models with non-Gaussian priors due to the intractable marginal likelihood. Existing methods, including variational Bayes and Markov chain Monte Carlo (MCMC)-based approaches, often have high computational cost and slow convergence. We propose an efficient framework to approximate the Fisher information matrix (FIM) and the CRB by expressing the gradient of the log marginal likelihood as a posterior expectation. Expectation propagation (EP) is used to approximate the posterior as a Gaussian, enabling accurate moment estimation compared to pure sampling-based methods. Numerical experiments on small-scale sparse models show that the EP-based CRB approximation achieves lower average normalized mean squared error (NMSE) and faster convergence than classical baselines in non-Gaussian settings.
{"title":"Efficient CRB Estimation for Linear Models via Expectation Propagation and Monte Carlo Sampling","authors":"Fangqing Xiao;Dirk Slock","doi":"10.1109/LSP.2025.3648638","DOIUrl":"https://doi.org/10.1109/LSP.2025.3648638","url":null,"abstract":"The Cramér–Rao bound (CRB) quantifies the variance lower bound for unbiased estimators, but it is intractable to evaluate in linear hierarchical Bayesian models with non-Gaussian priors due to the intractable marginal likelihood. Existing methods, including variational Bayes and Markov chain Monte Carlo (MCMC)-based approaches, often have high computational cost and slow convergence. We propose an efficient framework to approximate the Fisher information matrix (FIM) and the CRB by expressing the gradient of the log marginal likelihood as a posterior expectation. Expectation propagation (EP) is used to approximate the posterior as a Gaussian, enabling accurate moment estimation compared to pure sampling-based methods. Numerical experiments on small-scale sparse models show that the EP-based CRB approximation achieves lower average normalized mean squared error (NMSE) and faster convergence than classical baselines in non-Gaussian settings.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"451-455"},"PeriodicalIF":3.9,"publicationDate":"2025-12-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929514","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As an emerging technology, Video Face Super-Resolution (VFSR) aims to reconstruct high-resolution facial images from low-quality video sequences while maintaining identity consistency, which makes it applicable to scenarios such as surveillance, video conferencing, and film restoration. Compared with image-based face restoration and general video super-resolution, VFSR is more challenging because it requires accurate facial detail reconstruction, strict identity preservation, and computational efficiency under varying poses and expressions. To address these challenges, we propose a High-Precision Identity Preserving VFSR framework (HPIP), which integrates a Multi-Scale Prediction Module (MPM) and an Identity Preservation Module (IPM). The MPM focuses on identity-critical facial regions (e.g., eyes, nose, and mouth) and leverages multi-scale feature prediction to improve reconstruction accuracy and robustness while maintaining computational efficiency. The IPM further projects features into a latent representation space, generating temporally consistent dictionary features and enhancing temporal coherence. Extensive experiments demonstrate that HPIP achieves superior performance in both qualitative and quantitative evaluations, producing visually pleasing facial details while maintaining an efficient and lightweight design.
视频人脸超分辨率（Video Face Super-Resolution， VFSR）是一项新兴技术，旨在从低质量的视频序列中重建高分辨率的人脸图像，同时保持身份一致性，适用于监控、视频会议和电影修复等场景。与基于图像的人脸恢复和一般的视频超分辨率相比，VFSR需要精确的人脸细节重建，严格的身份保持，以及不同姿态和表情下的计算效率，更具挑战性。为了解决这些问题，我们提出了一个高精度身份保持VFSR框架（HPIP），该框架集成了一个多尺度预测模块（MPM）和一个身份保持模块（IPM）。MPM专注于身份关键面部区域（例如，眼睛，鼻子和嘴巴），并利用多尺度特征预测来提高重建精度和鲁棒性，同时保持计算效率。IPM进一步将特征投射到潜在的表示空间中，生成时间一致的字典特征并增强时间一致性。大量的实验表明，HPIP在定性和定量评估中都取得了卓越的性能，在保持高效和轻量化设计的同时，产生了视觉上令人愉悦的面部细节。
{"title":"Video Face Super-Resolution With High-Precision Identity Preservation","authors":"Chaoliang Wu;Ting Zhang;Xianbin Zhang;Nian He;Yiwen Xu","doi":"10.1109/LSP.2025.3648639","DOIUrl":"https://doi.org/10.1109/LSP.2025.3648639","url":null,"abstract":"As an emerging technology, Video Face Super-Resolution (VFSR) aims to reconstruct high-resolution facial images from low-quality video sequences while maintaining identity consistency, which makes it applicable to scenarios such as surveillance, video conferencing, and film restoration. Compared with image-based face restoration and general video super-resolution, VFSR is more challenging because it requires accurate facial detail reconstruction, strict identity preservation, and computational efficiency under varying poses and expressions. To address these challenges, we propose a High-Precision Identity Preserving VFSR framework (HPIP), which integrates a Multi-Scale Prediction Module (MPM) and an Identity Preservation Module (IPM). The MPM focuses on identity-critical facial regions (<italic>e.g.</i>, eyes, nose, and mouth) and leverages multi-scale feature prediction to improve reconstruction accuracy and robustness while maintaining computational efficiency. The IPM further projects features into a latent representation space, generating temporally consistent dictionary features and enhancing temporal coherence. Extensive experiments demonstrate that HPIP achieves superior performance in both qualitative and quantitative evaluations, producing visually pleasing facial details while maintaining an efficient and lightweight design.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"406-410"},"PeriodicalIF":3.9,"publicationDate":"2025-12-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929369","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Generative steganography has recently attracted considerable attention due to its superior security properties. However, most existing approaches suffer from limited hiding capacity. To address this issue, this paper proposes a high-capacity image steganography framework that integrates an encoder–decoder architecture with a latent diffusion model. Specifically, a message encoder is designed to transform binary secret messages into latent-space representations through a series of ResDense modules, enabling efficient hiding of large-scale information. The encoded latent features are then guided by the latent diffusion model to synthesize visually realistic stego images. During message extraction, the stego image undergoes iterative noise addition within the diffusion process to reconstruct the latent representation, from which a message decoder accurately recovers the hidden message. Extensive experimental results demonstrate that the proposed method achieves a high hiding capacity of over 30,000 bits, outperforming state-of-the-art methods while ensuring reliable message recovery under common image storage formats such as JPEG and PNG.
{"title":"High-Capacity Image Steganography via Latent Diffusion Models","authors":"Ruijie Du;Na Wang;Cheng Xiong;Chuan Qin;Xinpeng Zhang","doi":"10.1109/LSP.2025.3647567","DOIUrl":"https://doi.org/10.1109/LSP.2025.3647567","url":null,"abstract":"Generative steganography has recently attracted considerable attention due to its superior security properties. However, most existing approaches suffer from limited hiding capacity. To address this issue, this paper proposes a high-capacity image steganography framework that integrates an encoder–decoder architecture with a latent diffusion model. Specifically, a message encoder is designed to transform binary secret messages into latent-space representations through a series of ResDense modules, enabling efficient hiding of large-scale information. The encoded latent features are then guided by the latent diffusion model to synthesize visually realistic stego images. During message extraction, the stego image undergoes iterative noise addition within the diffusion process to reconstruct the latent representation, from which a message decoder accurately recovers the hidden message. Extensive experimental results demonstrate that the proposed method achieves a high hiding capacity of over 30,000 bits, outperforming state-of-the-art methods while ensuring reliable message recovery under common image storage formats such as JPEG and PNG.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"401-405"},"PeriodicalIF":3.9,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929556","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-23DOI: 10.1109/LSP.2025.3647431
Lizhuo Liu;Songbo Wang;Yimin Fu
Existing infrared small target detection (IRSTD) methods typically assume that training and testing data share the same distribution. However, this assumption often fails in real-world applications due to environmental and sensor-induced variations, resulting in significant performance degradation caused by domain shifts. Besides, the inherently low signal-to-clutter ratio of targets in infrared images further impedes the extraction of underlying target information, increasing the risk of overfitting to domain-specific patterns. This severely constrains the generalizability of knowledge learned from source domains, particularly when training is confined to a single source domain due to the high cost of data annotation. To solve this problem, we propose hierarchical structure dependency whitening (HSDW) for single-domain generalized IRSTD. Specifically, we characterize domain discrepancies in infrared images as differences in structural information. Building upon this point, we employ feature whitening to mitigate the dependency on domain-specific structure information, whose distribution is diversely simulated by a dual-branch nonlinear transformation module. Moreover, we adopt a hierarchical suppression mechanism to alleviate the structural biases across multiple decoding stages, thereby facilitating more generalized target understanding across domains. Extensive experiments on three public IRSTD datasets demonstrate that our method achieves state-of-the-art performance.
{"title":"Hierarchical Structure Dependency Whitening for Single-Domain Generalized Infrared Small Target Detection","authors":"Lizhuo Liu;Songbo Wang;Yimin Fu","doi":"10.1109/LSP.2025.3647431","DOIUrl":"https://doi.org/10.1109/LSP.2025.3647431","url":null,"abstract":"Existing infrared small target detection (IRSTD) methods typically assume that training and testing data share the same distribution. However, this assumption often fails in real-world applications due to environmental and sensor-induced variations, resulting in significant performance degradation caused by domain shifts. Besides, the inherently low signal-to-clutter ratio of targets in infrared images further impedes the extraction of underlying target information, increasing the risk of overfitting to domain-specific patterns. This severely constrains the generalizability of knowledge learned from source domains, particularly when training is confined to a single source domain due to the high cost of data annotation. To solve this problem, we propose hierarchical structure dependency whitening (HSDW) for single-domain generalized IRSTD. Specifically, we characterize domain discrepancies in infrared images as differences in structural information. Building upon this point, we employ feature whitening to mitigate the dependency on domain-specific structure information, whose distribution is diversely simulated by a dual-branch nonlinear transformation module. Moreover, we adopt a hierarchical suppression mechanism to alleviate the structural biases across multiple decoding stages, thereby facilitating more generalized target understanding across domains. Extensive experiments on three public IRSTD datasets demonstrate that our method achieves state-of-the-art performance.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"396-400"},"PeriodicalIF":3.9,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929372","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-19DOI: 10.1109/LSP.2025.3646141
Saidur R. Pavel;Yimin D. Zhang
This letter considers two-dimensional direction of arrival (DOA) estimation of coherent signals exploiting a moving uniform rectangular array. The motion of the array induces phase variations in the received signals across spatial positions, enabling the construction of decorrelated covariance matrices through forward-backward spatial smoothing. We analyze the achievable degrees of freedom (DOFs) in terms of movement steps and examine the impact of the motion support on effective decorrelation. Notably, we show that the maximum number of DOFs can be achieved if each movement step is at least half the signal wavelength and the number of movement steps is no less than half the number of array elements. Furthermore, it is demonstrated that distributing motion across both array axes yields better decorrelation and estimation performance than restricting movement to a single dimension.
{"title":"2D DOA Estimation of Coherent Signals Exploiting Moving Uniform Rectangular Array","authors":"Saidur R. Pavel;Yimin D. Zhang","doi":"10.1109/LSP.2025.3646141","DOIUrl":"https://doi.org/10.1109/LSP.2025.3646141","url":null,"abstract":"This letter considers two-dimensional direction of arrival (DOA) estimation of coherent signals exploiting a moving uniform rectangular array. The motion of the array induces phase variations in the received signals across spatial positions, enabling the construction of decorrelated covariance matrices through forward-backward spatial smoothing. We analyze the achievable degrees of freedom (DOFs) in terms of movement steps and examine the impact of the motion support on effective decorrelation. Notably, we show that the maximum number of DOFs can be achieved if each movement step is at least half the signal wavelength and the number of movement steps is no less than half the number of array elements. Furthermore, it is demonstrated that distributing motion across both array axes yields better decorrelation and estimation performance than restricting movement to a single dimension.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"346-350"},"PeriodicalIF":3.9,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145879962","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This work proposes a physics-guided unified deep learning architecture for single image restoration targeting dehazing, deraining, and low-light enhancement. The architecture first estimates the transmission map and airlight under an atmospheric scattering model, and then refines the result with a grayscale prior. A DiffConv feature extractor is proposed which combines vanilla and difference convolutions with a Laplacian branch (to capture high-frequency features). During inference, its branches are re-parameterized into a single kernel for reducing computational complexity. The grayscale prior replaces the Y channel in the YCbCr space to suppress noise and color artifacts, while a refinement stage uses Spatial Feature Transform (SFT) to inject structural features from this grayscale prior into the RGB domain. Experiments on standard benchmarks show consistent improvements in PSNR and SSIM at lower computational cost.
{"title":"ASM-DiffConvNet: Physics-Guided Difference Convolution Network for Single-Image Restoration","authors":"Hemkant Nehete;Amit Monga;Tharun Kumar Reddy Bollu;Balasubramanian Raman","doi":"10.1109/LSP.2025.3646138","DOIUrl":"https://doi.org/10.1109/LSP.2025.3646138","url":null,"abstract":"This work proposes a physics-guided unified deep learning architecture for single image restoration targeting dehazing, deraining, and low-light enhancement. The architecture first estimates the transmission map and airlight under an atmospheric scattering model, and then refines the result with a grayscale prior. A DiffConv feature extractor is proposed which combines vanilla and difference convolutions with a Laplacian branch (to capture high-frequency features). During inference, its branches are re-parameterized into a single kernel for reducing computational complexity. The grayscale prior replaces the Y channel in the YCbCr space to suppress noise and color artifacts, while a refinement stage uses Spatial Feature Transform (SFT) to inject structural features from this grayscale prior into the RGB domain. Experiments on standard benchmarks show consistent improvements in PSNR and SSIM at lower computational cost.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"336-340"},"PeriodicalIF":3.9,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886556","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-19DOI: 10.1109/LSP.2025.3646140
Haijun Wang;Wei Hao;Lihua Qi;Haoyu Qu;Zihao Su
Visual object tracking methods utilizing onboard cameras have significantly advanced the widespread application of uncrewed aerial vehicles (UAVs). However, the stochastic and intricate noise inherent in camera systems has critically impeded the performance of UAV trackers, particularly under low-light conditions. To solve this problem, this letter presents an efficient wavelet-based denoising transformer (WTM) integrated with a fast Fourier adjustment module (FFAM) to reduce random real noise, thereby improving UAV nighttime tracking performance. Specifically, an encoder-latent-decoder structure is designed for efficient end-to-end transformation. Additionally, the WTM in both the encoder and the decoder block introduces channel-wise transformer to extract low-frequency information. The FFAM is utilized in the latent block to adjust local texture details. Finally, a novel residual feedforward network is designed to enhance the processing of high-frequency information. Extensive experimental results validate the effectiveness of our proposed method, demonstrating significant improvements in UAV nighttime tracking capabilities by adapting to diverse enhancers and tracking algorithms.
{"title":"Wavelet-Based Denoising Transformer With Fourier Adjustment for UAV Nighttime Tracking","authors":"Haijun Wang;Wei Hao;Lihua Qi;Haoyu Qu;Zihao Su","doi":"10.1109/LSP.2025.3646140","DOIUrl":"https://doi.org/10.1109/LSP.2025.3646140","url":null,"abstract":"Visual object tracking methods utilizing onboard cameras have significantly advanced the widespread application of uncrewed aerial vehicles (UAVs). However, the stochastic and intricate noise inherent in camera systems has critically impeded the performance of UAV trackers, particularly under low-light conditions. To solve this problem, this letter presents an efficient wavelet-based denoising transformer (WTM) integrated with a fast Fourier adjustment module (FFAM) to reduce random real noise, thereby improving UAV nighttime tracking performance. Specifically, an encoder-latent-decoder structure is designed for efficient end-to-end transformation. Additionally, the WTM in both the encoder and the decoder block introduces channel-wise transformer to extract low-frequency information. The FFAM is utilized in the latent block to adjust local texture details. Finally, a novel residual feedforward network is designed to enhance the processing of high-frequency information. Extensive experimental results validate the effectiveness of our proposed method, demonstrating significant improvements in UAV nighttime tracking capabilities by adapting to diverse enhancers and tracking algorithms.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"341-345"},"PeriodicalIF":3.9,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886624","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-19DOI: 10.1109/LSP.2025.3646139
Yu Sun;Rui Zhou;Yueli Cui;Ming Jin;Gangyi Jiang
The limited field of view (FoV) of commercial light field cameras has driven development of various stitching techniques to generate wide-FoV light field images (WLFIs). However, these techniques often introduce local distortions and angular inconsistencies, posing significant challenges for WLFI quality assessment. In this letter, a no-reference WLFI quality assessment (WLFIQA) method based on structured representation and progressive learning is proposed. Specifically, considering the high-dimensional characteristics and distortion properties of WLFIs, a novel joint spatial–angular representation strategy is first designed. For the angular domain, horizontal and vertical sub-aperture images stacks are employed to characterize angular features; for the spatial domain each sub-aperture image is divided into four quadrants, and image blocks containing complementary cues from corresponding positions across different quadrants are used as input for subsequent feature extraction. Furthermore, a global–local feature extraction network is employed to further model multi-scale distortion characteristics. Finally, a progressive learning strategy is designed to enhance the performance of overall perceptual evaluation. Experimental results on a benchmark WLFI dataset show that the proposed method outperforms existing quality methods.
{"title":"No-Reference Stitched Wide Field of View Light Field Image Quality Assessment via Structured Representation and Progressive Learning","authors":"Yu Sun;Rui Zhou;Yueli Cui;Ming Jin;Gangyi Jiang","doi":"10.1109/LSP.2025.3646139","DOIUrl":"https://doi.org/10.1109/LSP.2025.3646139","url":null,"abstract":"The limited field of view (FoV) of commercial light field cameras has driven development of various stitching techniques to generate wide-FoV light field images (WLFIs). However, these techniques often introduce local distortions and angular inconsistencies, posing significant challenges for WLFI quality assessment. In this letter, a no-reference WLFI quality assessment (WLFIQA) method based on structured representation and progressive learning is proposed. Specifically, considering the high-dimensional characteristics and distortion properties of WLFIs, a novel joint spatial–angular representation strategy is first designed. For the angular domain, horizontal and vertical sub-aperture images stacks are employed to characterize angular features; for the spatial domain each sub-aperture image is divided into four quadrants, and image blocks containing complementary cues from corresponding positions across different quadrants are used as input for subsequent feature extraction. Furthermore, a global–local feature extraction network is employed to further model multi-scale distortion characteristics. Finally, a progressive learning strategy is designed to enhance the performance of overall perceptual evaluation. Experimental results on a benchmark WLFI dataset show that the proposed method outperforms existing quality methods.","PeriodicalId":13154,"journal":{"name":"IEEE Signal Processing Letters","volume":"33 ","pages":"391-395"},"PeriodicalIF":3.9,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","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：