SAIGFormer: A Spatially-Adaptive Illumination-Guided Network for Low-Light Image Enhancement

2025-07-21

AI Summary

Existing Transformer-based methods for low-light enhancement often suffer from overexposure or insufficient brightness recovery under non-uniform illumination (e.g., backlighting, shadows). To address this, we propose SAIGFormer, a spatially adaptive illumination-guided network. Its core innovations are: (1) the first integration of dynamic integral image modeling into Transformers to enable precise, pixel-wise estimation of spatially varying illumination; and (2) illumination-guided multi-head self-attention (IG-MSA), which leverages illumination priors to modulate attention weights and emphasize brightness-sensitive regions. We further introduce the SAI²E module for end-to-end illumination-aware feature calibration. Extensive experiments on five standard low-light datasets and the cross-domain LOL-Blur benchmark demonstrate that SAIGFormer significantly outperforms state-of-the-art methods, particularly excelling in non-uniform scenarios with superior detail fidelity, natural luminance distribution, and cross-domain generalization capability.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Recent Transformer-based low-light enhancement methods have made promising progress in recovering global illumination. However, they still struggle with non-uniform lighting scenarios, such as backlit and shadow, appearing as over-exposure or inadequate brightness restoration. To address this challenge, we present a Spatially-Adaptive Illumination-Guided Transformer (SAIGFormer) framework that enables accurate illumination restoration. Specifically, we propose a dynamic integral image representation to model the spatially-varying illumination, and further construct a novel Spatially-Adaptive Integral Illumination Estimator ($ ext{SAI}^2 ext{E}$). Moreover, we introduce an Illumination-Guided Multi-head Self-Attention (IG-MSA) mechanism, which leverages the illumination to calibrate the lightness-relevant features toward visual-pleased illumination enhancement. Extensive experiments on five standard low-light datasets and a cross-domain benchmark (LOL-Blur) demonstrate that our SAIGFormer significantly outperforms state-of-the-art methods in both quantitative and qualitative metrics. In particular, our method achieves superior performance in non-uniform illumination enhancement while exhibiting strong generalization capabilities across multiple datasets. Code is available at https://github.com/LHTcode/SAIGFormer.git.

Problem

• Enhancing low-light images with non-uniform lighting issues
• Restoring accurate illumination in backlit and shadow scenarios
• Improving brightness restoration without over-exposure artifacts

Innovation

• Dynamic integral image representation for illumination modeling
• Spatially-Adaptive Integral Illumination Estimator (SAI2E)
• Illumination-Guided Multi-head Self-Attention (IG-MSA) mechanism