https://github.com/rishishanthan/rgb-channel-alignment
A Python project for reconstructing RGB images from grayscale plates with stacked channels using ORB feature matching and homography alignment.
https://github.com/rishishanthan/rgb-channel-alignment
channel-registration color-reconstruction computer-vision feature-matching homography image-alignment image-processing opencv orb python
Last synced: about 1 month ago
JSON representation
A Python project for reconstructing RGB images from grayscale plates with stacked channels using ORB feature matching and homography alignment.
- Host: GitHub
- URL: https://github.com/rishishanthan/rgb-channel-alignment
- Owner: rishishanthan
- Created: 2025-10-03T14:41:03.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2025-10-03T15:04:49.000Z (9 months ago)
- Last Synced: 2025-10-03T17:25:50.227Z (9 months ago)
- Topics: channel-registration, color-reconstruction, computer-vision, feature-matching, homography, image-alignment, image-processing, opencv, orb, python
- Language: Python
- Homepage:
- Size: 351 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# RGB Channel Alignment
This project reconstructs full-color RGB images from grayscale photographs where the Red, Green, and Blue channels are stacked vertically.
The implementation uses **ORB (Oriented FAST and Rotated BRIEF)** feature matching and **homography transformation** in OpenCV to align channels and correct misalignments, producing high-quality reconstructed images.
---
## 🎯 Objectives
- Split grayscale input (with 3 stacked channels) into individual **B**, **G**, and **R** channels.
- Use **feature-based alignment (ORB + Homography with RANSAC)** to correct channel misalignments.
- Warp and reconstruct aligned channels into a single RGB image.
- Save and visualize the reconstructed image.
---
## 🛠️ Implementation Details
### 1. **Splitting Channels**
Each raw input consists of three stacked grayscale images:
- Top third → Blue channel
- Middle third → Green channel
- Bottom third → Red channel
### 1. **Splitting Channels**
Each raw input consists of three stacked grayscale images:
- Top third → Blue channel
- Middle third → Green channel
- Bottom third → Red channel
### 2. Channel Alignment
- ORB (Oriented FAST and Rotated BRIEF) is used to detect and describe keypoints in each channel.
- BFMatcher with Hamming distance finds correspondences between feature descriptors.
- A homography is estimated with RANSAC to robustly align channels.
### 3. Reconstruction
- Red and Blue are aligned relative to Green (chosen as reference).
- Channels are stacked to form an RGB image.
### 4. Command-Line Interface
- The program accepts --input (raw stacked image) and --output (path to save aligned result).
- It now automatically handles directories or missing extensions.
### 5. Final Script Usage
Run the script as:
python3 untitled.py \
--input /tests/test_image_1.jpg \
--output /results/test_1_aligned.jpg
## ✅ Elample Result
#### Input - stacked grayscale plates
#### Output - aligned RGB image
## ⚙️ Dependencies
- Python 3.x
- OpenCV ≥ 4.5.4
- NumPy
- Matplotlib
### Install with:
pip install opencv-python numpy matplotlib
## 🔑 Key Takeaways
- This project demonstrates how early color photography relied on careful channel alignment.
- ORB + Homography with RANSAC gives robust alignment even with perspective distortions.
- Results show good alignment, though extreme motion blur or damaged plates may need advanced techniques (e.g., pyramid search, cross-correlation).
## 📌 Future Improvements
- Implement pyramid alignment (multi-scale search).
- Add automatic cropping to remove misaligned borders.
- Explore mutual information alignment for robustness.
## 📖 References
- S. Prokudin-Gorskii Photo Collection, Library of Congress
- OpenCV ORB and Homography Documentation