[Auto]Stitching Photo Mosaics

CS 194: Computational Photography, Spring 2020

Project 5

Abby Cohn

Part 1: Image Warping and Mosaicing

Overview

For the first part of this project, I rectified images using homographies. First, we define correspondences on the image we want to rectify but selecting at least 4 points. We select the correspondences on each pair of images and recover the transformation parameters between them. Thus, we can set up a linear system, aH = b, and solve for H using least-squares. We want to get H in the form of the 3X3 matrix on the right side of the equation below.

Since we should select more than 4 points for a more stable result, we use this equation below in the form ax = b and use least squares to get back H.

To actually warp the image, I used the inverse of H to transform the corresponding pixels in the image to the output image.