Dual-Path Learning based on Frequency Structural Decoupling and Regional-Aware Fusion for Low-Light Image Super-Resolution

Ji-Xuan He; Jia-Cheng Zhao; Feng-Qi Cui; Jinyang Huang; Yang Liu; Sirui Zhao; Meng Li; Zhi Liu

cs.CVMar 28, 2026

Dual-Path Learning based on Frequency Structural Decoupling and Regional-Aware Fusion for Low-Light Image Super-Resolution

Authors: Ji-Xuan He, Jia-Cheng Zhao, Feng-Qi Cui, Jinyang Huang, Yang Liu, Sirui Zhao, Meng Li, Zhi Liu

Organizations: Hefei University of Technology, Hefei, China · University of Science and Technology of China, Hefei, China · Zhejiang University, Hangzhou, China · The University of Electro-Communications, Tokyo, Japan

Abstract

Low-light image super-resolution (LLISR) is essential for restoring fine visual details and perceptual quality under insufficient illumination conditions with ubiquitous low-resolution devices. Although pioneer methods achieve high performance on single tasks, they solve both tasks in a serial manner, which inevitably leads to artifact amplification, texture suppression, and structural degradation. To address this, we propose Decoupling then Perceive (DTP), a novel frequency-aware framework that explicitly separates luminance and texture into semantically independent components, enabling specialized modeling and coherent reconstruction. Specifically, to adaptively separate the input into low-frequency luminance and high-frequency texture subspaces, we propose a Frequency-aware Structural Decoupling (FSD) mechanism, which lays a solid foundation for targeted representation learning and reconstruction. Based on the decoupled representation, a Semantics-specific Dual-path Representation (SDR) learning strategy that performs targeted enhancement and reconstruction for each frequency component is further designed, facilitating robust luminance adjustment and fine-grained texture recovery. To promote structural consistency and perceptual alignment in the reconstructed output, building upon this dual-path modeling, we further introduce a Cross-frequency Semantic Recomposition (CSR) module that selectively integrates the decoupled representations. Extensive experiments on the most widely used LLISR benchmarks demonstrate the superiority of our DTP framework, improving $+$ 1.6% PSNR, $+$ 9.6% SSIM, and $-$ 48% LPIPS compared to the most state-of-the-art (SOTA) algorithm. Codes are released at https://github.com/JXVision/DTP.

Figures & tables

Fig. 1: Comparison of existing methods and our proposed approach. (a) and (b) represent sequential and joint spatial-domain LLISR methods, respectively, which suffer from uneven luminance and blurred textures due to early-stage semantic entanglement. (c) illustrates our frequency-aware disentanglement strategy, which explicitly separates and models luminance and texture components, enabling faithful recovery with both crisp luminance and fine textures.

Fig. 2: An overview of the proposed DTP (a) The overall processing flow of DTP . (b) The pipeline of Decoupling Stage (FSD) in DTP . (c) The flow of Representation Specialization Stage (SDR). (d) The sketch of Fusion and Reconstruction Stage (CSR).

Fig. 3: Qualitative comparisons of $\times$ 2 tasks on RELLISUR. The top row shows the full restored images, the middle displays zoomed-in patches for local structure inspection, and the bottom presents RGB histogram distributions.

Fig. 4: Visual comparison under extreme low-light conditions ( $-2.5$ EV to $-4.5$ EV) on the RELLISUR dataset.

Method	Years	RELLISUR $\times$ 2			RELLISUR $\times$ 4
Method	Years	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
SwinIR [ 10 ]	ICCV’21	18.38 ↑4.77	0.640 ↑0.176	0.577 ↓0.453	17.53 ↑4.33	0.663 ↑0.151	0.688 ↓0.435
Restormer [ 15 ]	CVPR’22	21.21 ↑1.94	0.727 ↑0.089	0.385 ↓0.261	20.29 ↑1.57	0.720 ↑0.094	0.492 ↓0.239
OmniSR [ 12 ]	CVPR’23	20.87 ↑2.28	0.740 ↑0.076	0.470 ↓0.346	19.89 ↑1.97	0.723 ↑0.091	0.461 ↓0.208
HAT [ 2 ]	CVPR’23	20.21 ↑2.94	0.719 ↑0.097	0.454 ↓0.330	19.75 ↑2.11	0.715 ↑0.099	0.561 ↓0.308
PDHAT+ [ 9 ]	T-MM’24	-	-	-	21.00 ↑0.86	0.753 ↑0.061	0.452 ↓0.159

TABLE I: Comparison of our DTP with sota methods on RELLISUR dataset at

\times

2 and

\times

4 scales. ( Bold : Best, Underline : Second best)

Scale	Modules			Metric
	FSD	SDR	CSR	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
$\times 2$	$\times$	$\times$	$\times$	18.51	0.765	0.387
	✓	$\times$	$\times$	22.36 ↑3.85	0.803 ↑0.038	0.291 ↓0.096
	$\times$	✓	$\times$	20.27 ↑1.76	0.782 ↑0.017	0.362 ↓0.025
	$\times$	$\times$	✓	21.79 ↑3.28	0.791 ↑0.026	0.313 ↓0.074
	✓	✓	$\times$	22.75 ↑4.24	0.811 ↑0.046	0.203 ↓0.184

TABLE II: Ablation study of FSD, SDR, and CSR modules in DTP across

\times 2

and

\times 4

upscaling factors.

Explore similar work

Jun 27, 2026

FreqOrtho-SR: Frequency-Guided Orthogonal Expert Learning for Real-World Image Super-Resolution

Diffusion prior-based methods have shown impressive results in real-world image super-resolution (ISR), yet two key challenges persist: balancing pixel-level fidelity with semantic quality, and adapting to diverse degradations. Existing dual-branch approaches freeze the pixel module during semantic training, but the semantic branch can still expand capacity within the pixel subspace, precluding genuine perceptual improvement. Moreover, using a single static adapter cannot generalize across heterogeneous real-world corruptions. To address both issues, we propose FreqOrtho-SR, which comprises:

\textbf{Freq}

uency-guided Mixture of LoRA Experts (FreqMoE), it routes inputs to specialized experts via a non-parametric FFT-based degradation-feature extractor that encodes frequency-domain signatures, enabling stable and interpretable specialization across corruption types; and

\textbf{Ortho}

gonal Gradient Projection (OGP), which reframes the dual-objective optimization as a subspace-constrained problem: by extracting the pixel-fidelity subspace via SVD on combined expert weight deltas and projecting semantic gradients onto its null space, OGP guarantees orthogonality between the two objectives, enabling genuinely complementary learning without mutual interference. Experiments show that FreqOrtho-SR achieves competitive overall performance and a strong fidelity-perception trade-off across multiple benchmarks with efficient single-step inference. The source code of our method can be found at sonhm3029/FreqOrtho-SR.

Minh Son Hoang, Dinh Phu Tran, Quyen Nguyen Duc +2

School of Computing, KAIST, Republic of Korea

Diffusion Models for Image Restoration Mixture of Experts Image Super-Resolution

Ask PDF

Jul 27, 2026

Low-light Image Enhancement via Multi-scale Attention combined with Fourier Transform

Low-light image enhancement (LLIE) aims to improve image quality and clarity in diverse and demanding low-illumination environments. However, existing deep learning-based LLIE methods struggle to accurately capture real-world illumination and restore texture details, largely because their algorithmic strengths remain underutilized. To address these issues, we present a supervised frequency domain deep learning network for LLIE, named multi-scale attention combined with the Fourier transform (MSFT) which adopts a U-shaped, one-stage architecture that infuses guidance from low-light images into the network by channeling it through multi-scale attention. We further fuse the amplitude information from priori channels with that of the low-light image in MSFT's self-created module, and carry out multi-scale guidance along with the network. Subsequently, to better enhance the faint feature, such as fine content and textures, and to better fuse global context confidence in the decoding stage, we separately introduce a multi-shape synergistic attention and a lightweight network that effectively integrate information in high-dimensional space to embed into the superlative feature space channel containing rich texture information. Extensive experiments conducted on LOL, SID, SMID, and SDSD datasets demonstrate that MSFT significantly outperforms state-of-the-art competitors. For example, compared with Retinexformer, our method achieves a peak signal-to-noise ratio of up to 41.76 decibels on the SDSD-outdoor dataset with an increase of 11.92 decibels and a structural similarity index of 0.988 with a 13.80% improvement.

Wenbin Du, Jian Long, Zhu Cao

aKey Laboratory of Smart Manufacturing in Energy Chemical Process, Ministry of Education, East China University of Science and Technology, Shanghai, 200237, China · bCollege of Electronics and Information Engineering, Tongji University, Shanghai 201804, China · cShanghai Research Institute for Intelligent Autonomous Systems, Tongji University, Shanghai 201210, China

Frequency-Domain Feature Learning Low-Light Image Enhancement Multi-Scale Attention

Ask PDF

Apr 7, 2026

RHVI-FDD: A Hierarchical Decoupling Framework for Low-Light Image Enhancement

Low-light images often suffer from severe noise, detail loss, and color distortion, which hinder downstream multimedia analysis and retrieval tasks. The degradation in low-light images is complex: luminance and chrominance are coupled, while within the chrominance, noise and details are deeply entangled, preventing existing methods from simultaneously correcting color distortion, suppressing noise, and preserving fine details. To tackle the above challenges, we propose a novel hierarchical decoupling framework (RHVI-FDD). At the macro level, we introduce the RHVI transform, which mitigates the estimation bias caused by input noise and enables robust luminance-chrominance decoupling. At the micro level, we design a Frequency-Domain Decoupling (FDD) module with three branches for further feature separation. Using the Discrete Cosine Transform, we decompose chrominance features into low, mid, and high-frequency bands that predominantly represent global tone, local details, and noise components, which are then processed by tailored expert networks in a divide-and-conquer manner and fused via an adaptive gating module for content-aware fusion. Extensive experiments on multiple low-light datasets demonstrate that our method consistently outperforms existing state-of-the-art approaches in both objective metrics and subjective visual quality.

Junhao Yang, Bo Yang, Hongwei Ge +3

Jilin University, Changchun, China · Dalian University of Technology, Dalian, China · National University of Singapore, Singapore

Low-Light Image Enhancement Image Denoising Frequency-Domain Feature Learning

Ask PDF