Low Light Image Enhancement Using Transfer Learning on a Lightweight U-Net Architecture

Abstract

Low-Light Image Enhancement (LLIE) plays a crucial role in photography, surveillance, autonomous systems, and scientific imaging. Traditional enhancement techniques often struggle to recover fine details and may introduce unwanted color distortions. In this project, a lightweight U-Net model will be developed for low-light image enhancement. Transfer learning will be employed by first pretraining the model on a synthetic low-light dataset, followed by fine-tuning on real paired low-light images. Using the LOw-Light (LOL) dataset on Kaggle, model performance will be evaluated before and after transfer learning using Peak Signal-to Noise Ratio (PSNR) and Structural Similarity Index (SSIM) image quality metrics. The lightweight design of the model makes it well-suited for deployment on edge devices and mobile platforms. Keywords ⎯ Artificial Intelligence, Computer Vision, Edge ML, Low Light Image Enhancement, Low-Light Dataset, Transfer Learning, U-Net Architecture.