KAN See In the Dark

2024-09-05

arXiv.org

AI Summary

Low-light image enhancement faces challenges in modeling complex nonlinear mappings due to non-uniform illumination and noise corruption. To address this, we introduce the Kolmogorov–Arnold Network (KAN)—for the first time in low-level vision—proposing the KAN-Block module, which employs learnable spline-based activation functions for highly expressive nonlinear fitting. We further integrate spatial-frequency joint modeling with an inverse-diffusion-guided frequency-aware mechanism to enhance interpretability and robustness. Extensive experiments demonstrate state-of-the-art performance on mainstream benchmarks including LOL and SID, significantly outperforming CNN- and Transformer-based baselines. The proposed method achieves superior visual quality and quantitative metrics (e.g., PSNR, SSIM), while maintaining computational efficiency. All source code is publicly available to foster reproducibility and community advancement.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Existing low-light image enhancement methods are difficult to fit the complex nonlinear relationship between normal and low-light images due to uneven illumination and noise effects. The recently proposed Kolmogorov-Arnold networks (KANs) feature spline-based convolutional layers and learnable activation functions, which can effectively capture nonlinear dependencies. In this paper, we design a KAN-Block based on KANs and innovatively apply it to low-light image enhancement. This method effectively alleviates the limitations of current methods constrained by linear network structures and lack of interpretability, further demonstrating the potential of KANs in low-level vision tasks. Given the poor perception of current low-light image enhancement methods and the stochastic nature of the inverse diffusion process, we further introduce frequency-domain perception for visually oriented enhancement. Extensive experiments demonstrate the competitive performance of our method on benchmark datasets. The code will be available at: https://github.com/AXNing/KSID}{https://github.com/AXNing/KSID.

Problem

• Enhancing low-light images with nonlinear relationships
• Addressing illumination and noise in image enhancement
• Improving interpretability in low-level vision tasks

Innovation

• KAN-Block for image enhancement
• Frequency-domain perception introduced
• Spline-based convolutional layers utilized