Low-Light Image Enhancement with Deep Learning

Overview

This project focuses on developing a deep learning model that enhances images taken in low-light conditions while preserving important details and avoiding over-exposure artifacts.

The Problem

Images captured in low-light environments suffer from:

Our Approach

We developed a CNN-based architecture that:

  1. Analyzes the illumination map of the input image
  2. Separates noise from actual image content
  3. Enhances brightness while preserving natural colors
  4. Restores fine details lost in dark regions

Key Technologies

Results

Our model achieved:

Published at IJCNN 2025 as “RSEND: Retinex-based Squeeze-and-Excitation Network with Dark channel prior”

Before/After Examples

The model successfully enhanced cathedral interiors, street scenes, and indoor photography while maintaining natural color balance and avoiding the “washed out” appearance common in traditional enhancement methods.

Future Work

View Paper | GitHub Repository