Chapter 8: Image Alignment and Stitching
Pairwise alignment, image stitching, global alignment, and compositing methods.
Chapter Overview
Feature matching tells us which points correspond—but now what? We need to compute the actual transformation that aligns one image to another. This chapter covers the geometry and algorithms for image alignment.
What is this chapter about? We learn to compute geometric transformations from matched features, handle the inevitable mismatches with RANSAC, and combine multiple images into seamless panoramas.
Why does this matter? Image alignment enables:
- Panorama photography: Stitching together multiple photos
- Document scanning: Rectifying perspective-distorted documents
- Medical imaging: Aligning scans from different times or modalities
- Augmented reality: Placing virtual objects on real surfaces
How the topics connect: We start with homographies—the transformation model for planar scenes. Then RANSAC handles outliers in our feature matches. Panorama stitching puts it all together, including blending to hide seams.
Chapter Roadmap
Click any topic to jump in
Homography & 2D Alignment
Projective transformations, DLT, the transformation hierarchy — the geometric model for planar alignment.
RANSAC
Robust estimation under outliers — random sampling, iteration bounds, adaptive termination.
Two routes to multi-image alignment
Panorama Stitching
Full stitching pipeline — warping, multi-band blending, and exposure compensation for seamless mosaics.
Bundle Adjustment
Global optimization of cameras and structure — Levenberg-Marquardt, Schur complement, robust costs.
Sign up to unlock this chapter
This chapter is part of PixelBank Premium. Create a free account, then upgrade to read the full lesson — concepts, walkthroughs, and exercises.