This is an overview of a few of the projects I did in a recent computer vision class.
Matlab source code available upon request.
Illumination Space
In his project, we combined a number of still images of the
same scene under different lighting conditions to create a set of 3
basis images that can be combined
to recreate any lighting condition. Below are the 3 basis images.
Homography
I created a homography (mapping from one image to another) with
360º of images of the Georgia Tech. campus. Below is the
final result. Here is the process:
- Use a feature detector (SIFT) to
find features in all images.
- Use RANSAC to find the best
matches between pairs of images.
- Create 4 homographies (one
for each face of a cube).
- Manually scale and shift the 4
images to eliminate seams in the cubicVR program.