🌄 Panorama Image Stitching - Computer Vision Project

This project builds a complete panorama image stitching pipeline from scratch, including feature matching, bundle adjustment, spherical warping, seam optimization, and advanced blending to generate high-quality panoramic images.

Just like how a phone camera creates a panorama, this project takes multiple overlapping images of the same scene and stitches them smoothly into a single wide image.

⚠️ Important

This project uses classical Computer Vision algorithms
❌ No Deep Learning
❌ No OpenCV automatic stitcher

---

🧩 What is Panorama Image Stitching?

Panorama stitching means:

• You take multiple images of the same scene
• Each image overlaps slightly with the next
• The system finds common points between images
• Images are aligned geometrically
• All images are blended into one large seamless image

---

🧠 How This Project Works

1️⃣ Input Images

• All images are stored inside the images/ folder
• Images are:
  • Taken from slightly different angles
  • Have overlapping regions

---

2️⃣ Feature Detection & Matching (features.py)

The first step is to find common points between images.

What happens:

• Important points (corners, edges) are detected using SIFT / MSOP
• Features are matched between image pairs
• Incorrect matches are removed using Lowe’s Ratio Test

👉 This step identifies which part of one image corresponds to another image.

---

3️⃣ Homography Estimation

Once matching points are found:

• A homography matrix is computed using RANSAC
• This matrix describes how one image should be transformed to align with another

👉 This step performs geometric alignment of images.

---

4️⃣ Bundle Adjustment (Bundle_Adj.py)

This is the most important and advanced part of the pipeline.

Problem:

• Small alignment errors accumulate when stitching many images

Solution:

• Bundle Adjustment optimizes all camera parameters together
• Uses Levenberg–Marquardt optimization
• Minimizes reprojection error between matched points

Optimized parameters:

• Camera rotation
• Camera focal length
• Camera center (principal point)

👉 This makes the panorama accurate, stable, and distortion-free.

---

5️⃣ Spherical Projection (main.py)

Why is this needed?

• Flat image stitching causes distortion for wide panoramas

What happens:

• Images are projected onto a virtual sphere
• Perspective distortion is reduced

👉 The panorama looks more natural and smooth.

---

6️⃣ Image Warping (blend.py)

Each image is:

• Warped into spherical coordinates
• Mapped to the correct position in panorama space
• Filled with transparent pixels where data is missing

---

7️⃣ Seam Finding (Graph Cut)

When images overlap:

• The system must decide where to cut and join images

This project uses:

• Graph Cut algorithm
• Finds seams where pixel differences are minimal

👉 This avoids visible stitching artifacts.

---

8️⃣ Image Blending

To hide seams and brightness differences, multiple blending techniques are used:

• Alpha Blending – Simple weighted averaging
• Laplacian Pyramid Blending – Multi-scale smooth blending
• Poisson Blending – Gradient-based seamless blending

👉 Final output looks like one single photograph.

▶️ How to Run the Project

Run the application by passing the path to the images folder and optional arguments.

Basic Usage

python main.py images

---

⭐ If this project helped you understand panorama stitching, please star the repository!

---