Daylight synthesis from nighttime images using ZRENet: a zero-reference CNN framework

View at publisher Summarise with VIRAT
Authors: Nikhil Kumar, Chirag Verma, Manasvi Gaur, Deval Verma, Neeraj Baghel
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Capturing clear and detailed images in low-light conditions remains a significant challenge across various domains. Traditional enhancement techniques often fall short, producing results with poor contrast and diminished detail. To address this, we introduce ZRENet, a novel deep learning framework designed to enhance nighttime images without relying on paired training data or pre-trained models. ZRENet employs a parameter prediction network to estimate key enhancement factors—such as brightness, contrast, and midtone levels—and a dynamic weighting mechanism that adapts these enhancements based on image content. A final sharpening stage using Unsharp Masking further refines image details. Experimental evaluations show that ZRENet improves visibility and feature clarity in low-light scenes, achieving an accuracy of 83% and outperforming existing methods by 5%. These results highlight ZRENet’s potential for real-world applications in surveillance, autonomous systems, and scientific imaging.

Fields of Study & Subjects

Image Enhancement Techniques Advanced Image Processing Techniques Image and Signal Denoising Methods

Publication details

Journal / source
International Conference on Optical and Wireless Technologies (OWT)
Pages
31–31
Publication date
30 March 2026
Language
EN