• Version
• Download 12
• File Size 429.65 KB
• File Count 1
• Create Date 20 March 2026
• Last Updated 20 March 2026

Deep Learning-Based Low-Light Image Enhancement for Improved Visibility

1 Katta. Vivek, 2 Eruguralla. Satish Babu, 3 Madha. Vaishnavi, 4 Vadluri. Srivalli, 5 Nagula. Rahul

Department of CSE (Artificial Intelligence & Machine Learning)

Jyothishmathi Institute of Technology and Science, Karimnagar, Telangana, India kattavivek40@gmail.com, vaishnavimadha1554@gmail.com, srivalli827@gmail.com, rahulnagula24@gmail.com

Abstract—Images captured under low light conditions usually have low visibility, noise, and a loss of fine details, making them less useful for real-world applications. Traditional methods of image enhancement often introduce color distortion or over- brightening of the images; meanwhile, the classic supervised deep learning approaches require paired datasets that are too challenging to obtain. This paper describes a deep learning-based low-light image enhancement system with Zero-Reference Deep Curve Estimation (Zero-DCE++). The proposed approach learns adaptive illumination correction functions directly from low-light inputs without requiring ground-truth reference images. Multiple no-reference loss functions shall guide the training process in order to preserve the natural color appearance of the image and assure spatial coherence, which will help in proper exposure control. The experimental results prove that the proposed method enhances visual clarity while preserving structural details and minimizes noise amplification.

Index Terms—Low-Light Image Enhancement, Self-Supervised Learning, Zero-DCE++, Deep Learning, Computer Vision