PanoStitch

A modular panoramic image stitching pipeline featuring both classic algorithms and modern Deep Learning integrations.

Overview

PanoStitch is an intelligent panoramic image stitching system that seamlessly combines multiple overlapping images into a single high-resolution panorama. Our modular architecture supports both traditional computer vision techniques and cutting-edge deep learning approaches, offering flexibility and robustness for various stitching scenarios.

Input Images

Image 1	Image 2

Result Panorama

Block Diagram

This diagram illustrates the complete pipeline architecture of our panoramic stitching system:

💡 Test Cases: You can find 8 comprehensive test cases with their stitched panorama results in the imgs/ folder (bicycle, boat, clock, dam, flower, mountain, river, tree).

Features

• Feature Detection: Custom Harris Corner Detector or SIFT.
• Feature Description: Custom 128D HOG-based descriptors.
• Matching: Vectorized Brute-force matcher with Lowe's ratio test.
• Deep Learning: Integration with DISK + LightGlue for robust matching.
• Homography: RANSAC with Direct Linear Transform (DLT).
• Processing: Gain compensation for exposure correction and weighted blending.

Installation

pip install -r requirements.txt

Usage

Basic Usage (Recommended)

Stitch all images in a directory using the standard SIFT pipeline:

python panostitch.py imgs/boat/

Comparing Methods

1. Custom "From Scratch" Implementation (Harris + HOG Descriptors)

python panostitch.py imgs/boat/ --harris

2. Deep Learning Pipeline (DISK + LightGlue)

python panostitch.py imgs/boat/ --dnn

Contributors

Amira Khalid	Youssef Noser	Hussein Mohammed	Abdulrahman Medhat