15-463 Project 4

Stitching Panoramas

Eric Trottier

2010-11-02

Part 1

Part 1 was fairly straightforward. I computed the homography with hand-selected points, and then blended the images together using the suggested "bwdist" as the alpha channel. Unfortunately, I can't seem to find my camera. Until I find it, I'm going to have to borrow some un-stitched photos from previous years.

Here are some results:

Rectifying:

Stitching:

And the most useful images ever:

Part 2

In this part of the assignment we attempt to do this stuff automatically. I still can't find my camera but I borrowed a friend's so I could try this baby on my own photos.
We start by finding the corners in an image:

Then find less using adaptive non-maximal suppression:

Then match two images together:

Then use RANSAC to get rid of outliers:

And from here we can use our old code to stitch the images together (images are found here).

The entire process looks like this:

% load images
im1 = loadScaled('me/P1010697s.jpg', 1);
im2 = loadScaled('me/P1010698s.jpg', 1);
im3 = loadScaled('me/P1010689s.jpg', 1);

% compute harris corners (includes ANMS)
pts1 = harris(im1);
pts2 = harris(im2);
pts3 = harris(im3);

% match points together
[r12,r21] = matchPts(im1, im2, pts1, pts2);
[r23,r32] = matchPts(im2, im3, pts2, pts3);

% attach image 1 (left) to image 2 (center)
[pan,shift] = stitch(im1, im2, r12, r21);

% pop the third one in the right spot
[pan2,shift2] = stitch(im3, pan, r32, ptTransform(r23, shift));

% at this point more images can be added,
% however it makes the panorama *huge* and
% MATLAB usually complains about memory
% (or maybe it's because I'm too cheap to buy more memory)

% to do this, one just transforms points in images to the same points
% in `pan2`. I do this above where the only transformation is shifting
% the points