Project DescriptionRectificationPanoramas
Project Description
Whenever we have two images taken from the same vantage point--that is to say, two images which share a center of projection-- there is a transformation that maps pixels in one image to pixels in another image. Such transformations are homographies.

Rectification

One application of homographies is rectification. We can transform an image so that four points in the image that share a plane in the scene become a rectangle in the image. The result is that the plane in the scene is viewed from straight on instead of in perspective.

Panoramas

If we have two or more images taken with the same center of projection, we can solve for a homography that maps one image plane onto the other. We can find this transformation by finding corresponding feature points in each image, which we can either specify manually, or infer automatically. Once we have correspondences, and homographies, we can stitch images together into panoramas.
Image Rectification
Panoramas

Hand-stitched

I created the following panoramas by manually selecting correspondence points between pairs of images.

Auto-stitched

These panoramas were created automatically, using a matlab program to find correspondences. The program first searches each image for feature points (using Harris corner detection), matches features in each image (using descriptors based on MOPS), then removes outliers (using RANSAC).