15463 Computational Photography
Project 4 - Stitching Photo Mosaics
Description of Project
This project was broken into two parts. The first part was to deal with image warping. The second part was to deal with feature matching and auto-stitching.
| My original images for the warping: | |
 |  |
| Alignment by hand | |
 | |
Part 1;
I started by computing the homographies according to the Heckbert paper. I computed H as the transform from source to destination. The computeH function currently only uses the 1st 4 correspondencies.
Next, I tried to get the image warping to work. Given a point in the source image, I multiplied it (as [x;y;1]) against H to calculate where I should place it into the warped image.
I've had varying success with the image warping. Unfortunately I didn't save all of the code versions I've used.
The first version was very "streaky" and had a lot of black lines through it. For this reason, I attempted to use interp2, but that took too long to generate a single image.
Here's the original warped image results (they have been manually aligned).
I next experimented with rectifying images, but discovered that there was an issue with my warp code. Here's a strange result:
Since then, I've worked on my warp code. This is the current result:
That's as far as I got. :(
