Project 2 Writeup

Tim St. Clair (tstclair)
February 19, 2010

Description

My implementation of the Poisson image blending technique is very similar to the method outlined in Pérez, et al., with the exception of edge cases. In my implementation, I allowed the neighboring pixels to wrap around the image. This simplified my algorithm significantly since I didn't need to special case the boundary cases, except for the top left and bottom right pixels which I did create a special case for. As a result of this simplification, I was able to speed up my implementation significantly, and was not able to see any visual difference.

Results

Here are some of the results of my implementation of the poisson image blending algorithm. In all the images, the naive blending is on the left for comparison, and the poisson blended image is on the right.

Default Images

My Images

This is my dad's face blended onto a picture of a grumpy looking Cardinal. I think in this case a gaussian pyramid approach to blending my actually work a litttle better, since the image boundary could be blurred more, and the colors are alreadf fairly close.
This is a picture of me on a glacial mountaineering trip, blended on to a picture of a distant glacier. The white snow gets blended into me a little more than I would have liked, but I liked how it made my head appear to be in the clouds. You can also see an example of the algorithm blurring the background around my legs.
This is my personal favorite. It is a picture of an egle diving to catch something blended with a picture of my friend in the mountains. The color blending is done well in this case, and because I made the mask fairly tight, there is not a lot of bluring that occurs.
Here is another case where the algorithm does very well, although this example is easy since it is blending water to water. The images are a lake in the mountains of Alaska, and some breeching Orcas that I found on the internet.
This is an example of where the algorithm does very poorly. What has happened here is that the image being blended in is on a dark background, and it is being blended over a patch of sky with a very light background. The result is the face gets overly whitened.