Image Warping and Mosaicing

Recovering Homographies

Our first objective was to recover the homography matrix, which defines the transformation between a pair of images. In our case, we want to find the transformation between our original input and our desired rectified output shape. To recover the matrix, we choose multiple correspondence points and solve using least squares.

Rectifying Images

The next step is to rectify the images. The trick here is to find sample images with planar surfaces that can be warped so that the plane is frontal-parallel. The next step is to define the correspondences for our desired output in rectangular coordinates, compute the homography, and warp our image. Below are some results of rectified artwork. I chose some famous paintings captured at unfortunate angles in an attempt to restore them. In the spirit of self-isolating, I rectified some lonely Edward Hopper paintings which are less lonely but also notably less pretty :P

Original Image	Rectified Image
starry night
frida kahlo
office in a small city
western motel