CS 194-26: Project 6

[Auto]Stitching Photo Mosaics

Part A: Image Warping and Mosaicing

Ellen Hong

Overview

In this project, we explore the applications of homographies in image warping and creating photo mosaics.

Image Warping

We demonstrate the results of warping images of the same scene as looked at from different perspectives. In order to create this effect, we apply a homographic transformation on the feature points from one image to another. The homography H is found by solving the following least-squares equation, where (xn, yn) is a feature point on one image, and (xn', yn') is the corresponding feature point on the other image.

Here are some examples of warped images:

Poster, side perspective
Poster, front perspective
Side perspective poster, warped to front perspective
Elevator sign, side perspective
Elevator sign, front perspective
Side perspective elevator sign, warped to front perspective

A not-as-successful result, due to the 3-D nature of the numbers:

Building number, side perspective
Building number, front perspective
Side perspective building number, warped to front perspective

Image Rectification

We can also use the same homographic transformation to rectify images, i.e. change the perspective of an image by manually defining the output feature points. We produce the following results:

Box, side perspective
Box, warped to front perspective
PG&E sign, side perspective
PG&E sign, warped to front perspective

Image Mosaicing

We can now use this warping algorithm to create mosaics, by warping point correspondences from one image to another, and blending the images together via linear alpha blending. We demonstrate this technique on the following results:

SF buildings, left view
SF bulidings, front view
SF buildings, right view
Blended mosaic, SF buildings
Rooftop view, left view
Rooftop view, front view
Rooftop view, right view
Blended mosaic, rooftop view
Living room, left view
Living room, front view
Living room, right view
Blended mosaic, living room