To select corners I used the starter code for computing harris corners. Then I applied adaptive non-maximal
suppression to select an even distribution of significant points.
I then extracted features by sub-sampling from the gaussian blurred image. I then paired the points based
on these features and thresholded them based on the ration of the first and second nearest neight; this
further reduced the number of points.
For the final mosaics, I used the same method of calculating homographies and merging the aligned images
as for part A. I think the results turned out very well except for the room panorama. I believe this did
not turn out as well because there were very few corners to detect due to the walls and there were many
false matches between points because the walls are very uniform.