This project is intended to explore stitching images together using matching features between images.
The intention is to have as little human interaction as possible. So the goal is to have the computer find correspondences between images.
If time permits, I would like to implement it for \(n\) images.
The program defines a class that stitches images together. At the command line the user gives a list of images and options for how to output the results (live or to disc).
python stitch.py img0.jpg img1.jpg --display --save output.jpg
When the images are provided the main function creates a Stitcher
class. From there the stitcher iterates through the images provided,
in order, and computes homography for the pair. It computes
homography by computing features in each image pair. These features
are computed by using the built in SIFT (Scale Invariant Feature
Transform). Once these features are found, the best features are
filtered into a list and this list is used. The user can specify the
correspondence threshold at the command line using the thresh
parameter. Once the best features are filtered we compute the
homography matrix using built in opencv functions. In order to
dynamically resize the output image, the algorithm computes the
conjunction of the canvas’ dimensions and the warped additional
image’s dimensions in order to determine the size of the new
canvas. Once the new canvas’ size is computed, the original image is
warped to be fit onto the new canvas, and the old canvas is placed
on top of that. Because this approach is iterative, it works for
many images in general, but the best results are when two images are
provided.