Panoramas by Feature matching

For this project I implemented a panorama-maker.

It uses simple square patch-appearance around corners as features and matches these features between consecutive images using RANSAC to get a consistent homography. The following panorama of the GRASP laboratory was created by stitching 10 images together:

![GRASP-panorama] (grasp_panorama_large.jpg)

As an extra-credit I implemented a couple of features to make the panoramas look natural:

1. 2-band blending of images for seamless transition as described in Lowe's paper Automatic Panoramic Image Stitching using Invariant Features.

2. Cylindrical warping of images before stitching. This prevents images at the end of the panorama from stretching too much. This approach is described in Szeliski's paper Creating Full View Panoramic Image Mosaics and Environment Maps. An example of this cylindrical warping is shown below:

![Cylinder-nowarp] (IMG_1062.JPG)
![Cylinder-warp] (cylinder.jpg)