View in EDS

A dual-guide aware normalizing flow for low-light enhancement with global illumination and cross-channel attention.

Saved in:
Bibliographic Details
Title: A dual-guide aware normalizing flow for low-light enhancement with global illumination and cross-channel attention.
Authors: Zhuang, Xue1 (AUTHOR) 2231221075@stu.xaut.edu.cn, Xiao, Zhaolin1 (AUTHOR) xiaozhaolin@xaut.edu.cn, Su, Haonan1 (AUTHOR) suhaonan@xaut.edu.cn
Source: Multimedia Systems. Jun2026, Vol. 32 Issue 4, p1-16. 16p.
Subjects: Image enhancement (Imaging systems), Luminosity, Night photography, Machine learning, Probabilistic generative models, Computer vision, Image processing
Abstract: Low Light Image Enhancement (LLIE) focuses on improving the quality of images captured under insufficient lighting conditions. Recently, normalizing flow model has been introduced to LLIE task due to its powerful capability in modeling data distribution. However, existing flow-based methods often overlook the importance of conditional features in guiding accurate probabilistic modeling, which can lead to uneven illumination and unnatural color restoration. To address this issue, we propose a Dual-Guide aware Normalizing Flow (DGNF) framework, which enhances encoding quality and injects richer conditional features into the flow-based decoder. Specifically, the proposed Dual-Guided aware Encoder Module (DGEM) consists of two specialized components–the Illumination-Guided Encoder Module (IGEM) and the Color-Guided Encoder Module (CGEM), which accurate models the features of illuminance and color of low light images respectively. The Illumination and Color aware Normalizing Flow Decoder (ICNFD) is designed to inject encoded features into the flow-based decoder, using them as conditional inputs to enable high-quality image enhancement. Experimental results on the LOL-V1 and LOL-V2 datasets show that our method achieves improvements of 1.36 dB in PSNR and 0.005 in SSIM compared to state-of-the-art methods, demonstrating the effectiveness of the proposed encoder in content preservation, noise reduction, and color restoration. [ABSTRACT FROM AUTHOR]
Copyright of Multimedia Systems is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Low Light Image Enhancement (LLIE) focuses on improving the quality of images captured under insufficient lighting conditions. Recently, normalizing flow model has been introduced to LLIE task due to its powerful capability in modeling data distribution. However, existing flow-based methods often overlook the importance of conditional features in guiding accurate probabilistic modeling, which can lead to uneven illumination and unnatural color restoration. To address this issue, we propose a Dual-Guide aware Normalizing Flow (DGNF) framework, which enhances encoding quality and injects richer conditional features into the flow-based decoder. Specifically, the proposed Dual-Guided aware Encoder Module (DGEM) consists of two specialized components–the Illumination-Guided Encoder Module (IGEM) and the Color-Guided Encoder Module (CGEM), which accurate models the features of illuminance and color of low light images respectively. The Illumination and Color aware Normalizing Flow Decoder (ICNFD) is designed to inject encoded features into the flow-based decoder, using them as conditional inputs to enable high-quality image enhancement. Experimental results on the LOL-V1 and LOL-V2 datasets show that our method achieves improvements of 1.36 dB in PSNR and 0.005 in SSIM compared to state-of-the-art methods, demonstrating the effectiveness of the proposed encoder in content preservation, noise reduction, and color restoration. [ABSTRACT FROM AUTHOR]
ISSN:09424962
DOI:10.1007/s00530-026-02241-w