夏靖远

Statistical Degradation State Modeling for Mask-free Shadow Removal Zhixiong Yang, Jingyuan Xia^*, Yuxin Zhao, Shengxi Li, Shuanghui Zhang, Zhen Liu, Yaowen Fu, Weidong Jiang, and Yongxiang Liu
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026

This paper represents shadows as physics-guided degradation states in the second-order statistical space of the Luminance-Variation domain. To address the non-orthogonal coupling between texture dispersion and illumination attenuation, an oblique decoupling mechanism extracts baseline attenuation and separates texture fluctuations from photometric uncertainty. The state representation is embedded in EM-LVNet, a deep-unrolled Generalized Expectation-Maximization architecture that alternates statistical prior inference with uncertainty-guided spatial restoration. Experiments demonstrate state-of-the-art mask-free restoration with compact computational overhead, superior cross-dataset generalization, and strong empirical inference idempotence.

Band-kernel Stochastic Learning for Unsupervised Blind Hyperspectral Image Super-Resolution Zhixiong Yang, Jingyuan Xia^*, Shengxi Li, Lingyu Zheng, Shuanghui Zhang, Li Liu, Yaowen Fu, Yongxiang Liu
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026
Paper / PDF / Code

This paper proposes BKSR, an unsupervised blind hyperspectral image super-resolution framework built on a unified BKX-HMM model. By jointly optimizing band selection, kernel estimation, and image restoration through stochastic band exploration, test-time kernel learning, and robust restoration, it effectively addresses unknown degradations while alleviating the conventional complexity-performance trade-off.

Blind Super-Resolution Via Meta-Learning and Markov Chain Monte Carlo Simulation
Jingyuan Xia, Zhixiong Yang^*, Shengxi Li, Shuanghui Zhang^*, Yaowen Fu, Deniz Gündüz, Xiang Li
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024
Paper / PDF / Code

This paper proposes a plug-and-play blind image super-resolution method based on the Meta-learning and Markov Chain Monte Carlo, which contributes to preventing bad local optimal solutions from the optimization perspective.

Metalearning-Based Alternating Minimization Algorithm for Nonconvex Optimization
Jingyuan Xia, Shengxi Li, Jun-Jie Huang^*, Zhixiong Yang, Imad M. Jaimoukha, Deniz Gündüz
IEEE Transactions on Neural Networks and Learning Systems, 2023
Paper / PDF / Code

This paper proposes a meta-learning-based alternating minimization (MLAM) method for nonconvex problems of multiple variables, which aims to minimize a part of the global losses over iterations instead of carrying minimization on each subproblem, and it tends to learn an adaptive strategy to replace the handcrafted counterpart resulting in advance on superior performance.

Meta-learning based Domain Prior with Application to Optical-ISAR Image Translation
Huaizhang Liao, Jingyuan Xia^*, Zhixiong Yang, Fulin Pan, Zhen Liu, Yongxiang Liu
IEEE Transactions on Circuits and Systems for Video Technology, 2023
Paper / PDF / Code

The proposed MLDP aims to realize the image translation from optical domain to ISAR domain. To generate realistic ISAR images, MLDP learns the scattering distribution features and the classification identifying features in physical and task perspectives. Meanwhile, MLDP applies the meta-learning strategy to improve its generalization ability to generate ISAR images under limited training samples.

Ultra-High-Definition Image Restoration via High Frequency Enhanced Transformer Chen Wu, Ling Wang, Zhuoran Zheng, Weidong Jiang, Yuning Cui^*, Jingyuan Xia
IEEE Transactions on Circuits and Systems for Video Technology, 2025
Paper / PDF

HiFormer is a dual-branch Transformer architecture designed for Ultra-High-Definition (UHD) image restoration. It combines a high-resolution branch that preserves fine details using directionally-sensitive large-kernel convolutions with a low-resolution branch that models global context via self-attention. This collaboration effectively compensates for high-frequency losses caused by downsampling and attention mechanisms, enabling efficient and high-fidelity restoration on consumer-grade GPUs.

Spectrally Spatially Coupled Dynamic Reduction for Efficient Hyperspectral Image Super Resolution Lingyu Zheng, Jingyuan Xia, Zhixiong Yang, Chen Wu, Shengxi Li, Xin Deng, Mai Xu^*
IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2026
Paper / PDF / Code

This paper proposes S$^2$CD-Net, a spectrally-spatially coupled dynamic lightweight network for hyperspectral image super-resolution. To address the high-dimensional redundancy and heavy computational cost of existing methods, S$^2$CD-Net adopts a coupled reduction and independent learning paradigm: spectral selection and spatial compression collaboratively reduce cross-dimensional redundancy, while lightweight SS2D scanners perform efficient global modeling on compact spectral and spatial sequences. Experiments show that S$^2$CD-Net achieves competitive or superior reconstruction quality with significantly fewer parameters and lower computational complexity, offering an effective performance-efficiency trade-off for edge deployment.

Efficient Image Dehazing via Prior-Guided Hierarchical Harmonization Network Chen Wu, Xiongfei Su, Yuning Cui, Mingyue Cui, Wenqi Ren, Jianhou Gan, Xin Yuan, Yulun Zhang, Jingyuan Xia^*
IEEE Transactions on Circuits and Systems for Video Technology, 2026
Paper

This paper proposes an efficient image dehazing network that embeds Dark Channel, Bright Channel, and Histogram Equalization priors as learnable operators rather than fixed shallow constraints. Through an asymmetric UNet with gated attention and dual bottlenecks, it hierarchically harmonizes physical priors with deep features, achieving state-of-the-art dehazing performance on 11 benchmarks with only 1.69M parameters and 17G FLOPs.

A Dynamic Kernel Prior Model for Unsupervised Blind Image Super-Resolution Zhixiong Yang, Jingyuan Xia^*, Shengxi Li, Xinghua Huang, Shuanghui Zhang, Zhen Liu, Yaowen Fu, Yongxiang Liu
CVPR, 2024
Paper / PDF / Code

The proposed DKP is a plug-and-play kernel estimation tool, which can be combined with the off-the-shelf image restoration model, e.g., DIP and Diffusion model, to realize unsupervised blind image super-resolution.

Luminance-Aware Statistical Quantization: Unsupervised Hierarchical Learning for Illumination Enhancement Derong Kong, Zhixiong Yang, Shengxi Li, Shuaifeng Zhi, Li Liu, Zhen Liu, Jingyuan Xia^*
NeurIPS, 2025
Paper / PDF / Code

LASQ reformulates low-light image enhancement as a statistical sampling process over hierarchical luminance distributions, leveraging a diffusion-based forward process to autonomously model luminance transitions and achieve unsupervised, generalizable light restoration across diverse illumination conditions.

Unlocking Optical Prior: Spectrum-Guided Knowledge Transfer for SAR Generalized Category Discovery Jingyuan Xia, Ruikang Hu, Ye Li, Zhixiong Yang, Xu Lan, Zhejun Lu
IEEE Transactions on Circuits and Systems for Video Technology, 2026
Paper / PDF

MCPT is a cross-modal pre-training framework that enables effective optical prior transfer for label-scarce SAR Generalized Category Discovery. This framework models cross-modal discrepancies in the frequency domain and employs discrepancy-aware feature refinement and contrastive alignment, improving SAR representation capability while achieving superior recognition performance across multiple mainstream datasets.

Dynamic Semantic Tokenization for Time Series via Elastic Sampling on Physics-aware Perception Huaizhang Liao, Zhixiong Yang, Jingyuan Xia^*, Yuheng Sun, Yue Zhang, Shengxi Li, Yongxiang Liu
AAAI, 2026
Paper / PDF / Code

PATK reformulates time series tokenization as a physics-aware semantic segmentation process over dual-domain signal distributions, leveraging a hidden Markov modeling mechanism to dynamically adapt token boundaries and achieve robust, interpretable representation learning across diverse temporal domains.

Scan Clusters, Not Pixels: A Cluster-Centric Paradigm for Efficient Ultra-high-definition Image Restoration Chen Wu, Ling Wang, Zhuoran Zheng, Yuning Cui, Zhixiong Yang, Xiangyu Chen, Yue Zhang, Weidong Jiang, Jingyuan Xia^*
CVPR, 2026
PDF / Code

This paper introduces C²SSM, a novel visual state space model for ultra-high-definition (UHD) image restoration. Unlike existing methods that scan every pixel, C²SSM adopts a cluster-centric paradigm: it first aggregates pixels into semantic clusters, then performs global modeling only on the cluster centroids, and finally diffuses the context back to all pixels. This significantly reduces computational cost while preserving high-quality restoration. Experiments on five UHD tasks show state-of-the-art performance with much lower complexity.