CS194 - Image Warping and Mosaicing

Recovering the Homography

To recover homographies from 2 sets of points, we need to solve for 3x3 matrix H with 8 unknowns. As such, we need at least 4 points to find H. To do this, we set up a system of linear equations and then solve it. For reference point sets greater than 4, we can implement least squares method. The recovered matrix H is then used to warp the image.

Shooting the Pictures

I took all pictures with my phone with fixed exposure and 50%-70% overlap region.

Image Warping and Rectification

After recovering the homography, we can rectify images and obtain the frontal view of planes such as building faces. This is done by choosing 4 points on rectangular building contours and then stretching them into an arbitrary rectangle. I used Evans Hall and the East Asian Library as examples. The images showcased here are cropped after being warped for better display.

Image Mosaicing

We can also use warping to create image mosaics similar to panoramas. I manually selected 6 points as references for each image on each junction. I used the middle image as an unwarped reference and aligned the left and right images to it to minimize pixel errors from manual reference points selection from adding up across the horizontal direction. The results are displayed below.