CS 194-26

IMAGE WARPING and MOSAICING

Randy Fan



Overview

In part A of the project, I took pictures, recovered homographies, warped images (with 2 rectified images as examples), and blended my own images into a mosaic.

Shoot and digitize pictures (20 pts)

Below are two images I took inside my house, and two images of museum galleries. They will be used for the below panoramas or rectifications.

Recover homographies (20 pts)

To recover homographies, I first recovered the homography vector using the matrix equation below. I then reshaped the vector into our desired 3x3 homography transformation matrix, where the last entry is 1 (scaling factor).

Warp the images (20 pts) [produce at least two examples of rectified images]

Below are two images and their rectified paintings. I obtained by warping (using the homography matrix) the portion of interest and specifying the ending shape (a rectangle in my examples)

Blend images into a mosaic (20 pts) [show source images and results for three mosaics.] (according to our TA, Shivam, just one panorama is enough for the checkpoint)

Below are two images I took, and, at the very bottom, their combined, contiguous image. I obtained the mosaic by warping one image using a homography transformation to be aligned with the other image, and then combining them with a simple blending technique (that specifies a lighter weight for pixels near the "center"/split of the two images.

Tell us what you've learned.

Homographies are very useful in panoramic photography, and selecting good corresponding points is important.