Monalisa's "smile"(minus 200px on both x and y)'
I basically implemented the baseline approach described in handout. Here is an outline for my algorithm
I retargeted all the given images to 200 px less than their original width and also retargeted the monalisa Image to 200px less than its original width and height. As for my own images, I retargetd them to 300 px less than their original width.
There are many failure cases. One of the most interesting one is the monolisa one showed on the right. At first glance, it appears to be ok, but if you look at it carefully, you will findout that the facial expression is actually different from the original images. All three of my own images also have some very apparent artifacts. I think those failure cases are caused by two major reasons. First, the algorithm has no idea of the spacial arrangement of the image. The second one is that if there are gradual change in the image, (like the moon light in one of my own image) the algorithm would fail to preserve those changes.
![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() |
![]() ![]() |
![]() ![]() |