In order to automatcally stitch a panorama, my algorithm used the following steps:
	1. Detects image features using the TA supplied Harris corner detection on two input images
	2. Uses Adaptive non-max suppression to lower the number of detected corners in both images to the same amount
	3. Collects image patches around each detection point and standardizes them.
	4. For each patch in one image it finds the most similar patch in the other images and if each patch is the other patch's most similar patch they are stored as a pair.
	5. These pairs are ranked according to the ratio of their similarity over the similarity between each patch and it's second most similar patch. This way the pairs which match with each other and no other patches are given the best ranking.
	6. The best 30 of these pairs are through the RANSAC algorithm to find a group of four that define a transformation that is correct for the most pairs.
	7. That group of 4 is used to create a transformation matrix and one image is warped using that matrix to hopefully fit on top of the other.
	8. Finally I crop the image into a retangle, so it looks better

Results Images

I took my pictures with my iphone so it at least partially explains why they weren't able to be stitched together that seamlessly.