Image Warping and Mosaicing

Project Members: Naren Krishna (cs194-26-abv)

Project Description

For this project, we first obtained photos (either via camera or online) of two scenes, with a 40% to 70% overlap between the views. From there, we calculated the parameters of a transformation between the two images, called a homography. We proceeded to do this by using the following equation to solve A * h = b, given pairs of corresponding points (x, y) and (x', y'):

We then proceeded, knowing the aforementioned homography, to warp one image into another and rectify the image so the plane is frontal-parallel by warping to a square. We did this by defining a point correspondence from the first images and manually inputting a point correspondence of a square as the second points. We then proceeded with the warp as indicated by the equation above. Finally, we used a weighted average combined with the warping technique to create the three image mosiacs shown below. As a bonus, the last image shows that image blending does not work as well when there is a disjoint part of the image, which is responsible for the unseemly cutoff when blended.

Rectified Images

Input Image #1 - Warping Entire Mural to Square (Online)

Rectified Image #1

Input Image #2 - Warping only Computer Screen to Square (My Camera)

Rectified Image #2

Mosaic Images

Input Image #1a (My Camera)

Input Image #1b (My Camera)

Mosaic Image #1

Input Image #2a (My Camera)

Input Image #2b (My Camera)

Mosaic Image #1

Input Image #3a (Online) - See explanation for unseemly line cutoff

Input Image #3b (Online)

Mosaic Image #3