CS 194-26 - (Auto)stitching and Photo Mosaics Part B

I first began the project by taking various images of the same scene at different angles. Here is an example…

Detecting Corner Features in an Image

I then used the given get_harris_corners function to find all the harris corners on the images and plotted them on the image. Originally, there were too many harris corners so to reduce this number, I applied the Adaptive Non-Maximal Suppression to reduce the number of corners.

We further reduced the number of harris corners by choosing a select number of harris corners to act as feature descriptors that are present in both images. After this we get something like this.

Using RANSAC to Compute a Homography & Creating Mosaic

Outdoor 1

Outdoor 2

Mosaic of outdoor 1 and outdoor 2

Closet 1

Closet 2

Mosaic of closet 1 and closet 2

Bathroom 1

Bathroom 2

Mosaic of bathroom 1 and bathroom 2

What I learned from this project

From this project, I learned how much we can do without training models. Before, I believed that techniques like stitching 2 images together would require extensive amount of data to train a model that would stitch these photos together. However, after this project, I realized that it is rather simple with the use of feature points to create homographies. I realized how powerful feature points are and how not everything with image processing requires machine learning.