CS 194-26 Project 6

Alvin Zhang

This project performs image stitching by computing homographies between images.


Manual

Image Rectification



I picked four points in a planar rectangle on the building for rectification. I then computed the homography from these points to a rectangle with corners at the center of each quadrant of the image, and used this homography to warp the original image.

Image Mosaic



By picking 6 feature points between pairs of images, I was able to compute the homography between them and stitch them together.

Automatic

Automatic Photo Stiching

Corner Detection

The first step is to find corners using the Harris corner detector.
Corner Non-Maximum Surpession

We then perform spatially-aware non-maximum surpression according to Lowe et. al.
Feature Matching

We extract features from the remaining corners and find the closest match between images.
RANSAC

We then use RANSAC to find the homography with the greatest number of inliers, and use this as our proposed homography.
Photo Stiching

Finally, we use the computed homography to warp and combine the images, as before.
Results