https://github.com/basharatwali/underwater-image-enhancement

https://github.com/basharatwali/underwater-image-enhancement

computer-vision image-enhancement image-processing underwater-images

Last synced: 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/basharatwali/underwater-image-enhancement
• Owner: BasharatWali
• License: mit
• Created: 2025-03-06T07:27:33.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2025-03-06T07:32:37.000Z (3 months ago)
• Last Synced: 2025-03-06T08:28:23.368Z (3 months ago)
• Topics: computer-vision, image-enhancement, image-processing, underwater-images
• Language: Jupyter Notebook
• Homepage:
• Size: 4.5 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Underwater Image Enhancement Pipeline

This repository contains an implementation of an underwater image enhancement pipeline. The project reproduces the DIRS-CLAHS method and extends it with several improvements to restore color balance and enhance local details in underwater images.

## Overview

Underwater images often suffer from low contrast, color imbalance, and haze due to the water medium. Our approach first reproduces the baseline DIRS-CLAHS method and then improves it by:

- Compensating for color attenuation (boosting the red channel)

- Refining global and local contrast correction

- Integrating a Multi-Scale Retinex module with color restoration

- Applying post-processing (gamma correction and weighted blending)

## Pipeline

The enhancement pipeline consists of the following steps:

- **Input:** Read the underwater image (and reference image if available for evaluation).

- **Color Attenuation Compensation (CAC):**  

  - Split the image into its B, G, and R channels.

  - Compute average intensities and derive a red compensation factor.

  - Boost the red channel while slightly attenuating the blue channel.

- **Global Contrast Correction:**  

  - Compute the 5th and 95th percentiles for each color channel.

  - Apply a piecewise linear stretch that maps the lower intensity region to [0, 127] and the higher region to [128, 255].

- **Adaptive Local Enhancement:**  

  - Convert the image to the LAB color space.

  - Apply CLAHE on the luminance (L) channel with an adaptive clip limit based on the image's brightness.

- **Multi-Scale Retinex with Color Restoration (MSRCR):**  

  - For each color channel, compute the Single-Scale Retinex (SSR) using Gaussian blurring with multiple scales.

  - Combine the SSR outputs using weighted averaging.

  - Apply a color restoration function (CRF) to mitigate desaturation.

- **Gamma Correction and Blending:**  

  - Apply mild gamma correction to adjust brightness.

  - Blend the gamma-corrected image with the locally enhanced image to achieve a balanced final output.

- **Evaluation:**  

  - Compute full-reference metrics (PSNR, SSIM, MSE) when a reference image is available.

  - Always compute no-reference metrics (UIQM, UCIQE) to assess visual quality.

## Evaluation Metrics

We use the following metrics:

- **PSNR:** Peak Signal-to-Noise Ratio

- **SSIM:** Structural Similarity Index

- **MSE:** Mean Squared Error

- **UIQM:** Underwater Image Quality Measure

- **UCIQE:** Underwater Color Image Quality Evaluation