Computational Photography

Zack Fleischman



Home
Project 0
Project 1
Project 2
Project 3
Project 4
Project 5


Project 1: Images
So I encountered some difficulties with the aligning in this lab. I used the Sum of Squared Differences method for aligning the color channels, but since the colors are not actually representative brightnesses and don't actually have to be correllated, it was not a very good method to do. I experimented with other methods such as normalized cross-correlation, but were unable to get better results. This method worked decently well for the small images, but it could not really home in on anything for the larger ones. The one thing that did work well is the pyramid images. I was able to only use a 3x3 area of search arounding a starting offset and then move that starting offset up for a higher resolution image. I used a gaussian blur and then halved the image to get the lower resolution images. This way, I was able to process the larger images, even though the alignments could have been better. Additionally, I also cropped the images somewhat since the edges of the images always had nasty color outliers.

//
The 3 Color ChannelsThe Color Image
00458u.tif Finished00458u.tif
01043u.tif Finished01043u.tif
01047u.tif Finished01047u.tif
01657u.tif Finished01657u.tif