CS 194-26 Project 6a: Image Warping & Mosaicing

Jennifer Liu (cs194-26-aag)

Shoot & Digitize Photos

Here are the original photos for my rectified images:

Here are the original photos for my mosaic blending images:

Recovering Homography: To recover the homography H, I needed a set of correspondence points between an image A and an image B. Using these, I could then define an equation Ah=B, and solve for the 8 variables in H using least squares.

Image Rectification: Here are the rectified images, achieved by computing the homography from the original images to a square (I estimated the points for a square), and warping the original image to its rectified equivalent using the homography recovered from the 4 correspondence points I selected from the original image and 4 points I used to define the vertices of a square to warp the image towards.

Mosaic

Here are my mosaic blended images:
For this part, I used the same warp function as in image rectification, but instead of warping an image to a square, I warped the image from the left side of a scene to the image depicting the right side of the same scene. After warping the left image to be in the same dimension / plane as the right image, I then used alpha blending with the distance transforms of the left and right images to blend the images together.

Takeaways: From this project, I learned that linear algebra, specifically perspective transforms, applied to images can create some interesting effects, as well as how difficult it is to choose correspondence features by hand. Looking forward to figuring out how to automate this process to spare myself from subpar warps due to human error :(.