A demonstration video. Not only is it a really interesting way to change image size, but tweaking the weights to remove objects very quickly in a way that at least at first glance looks natural is cool. Since they use gradient magnitude, I'm curious what technique they use when enlarging to prevent the inserted line being reused for further enlargement. It could be that it's important to always used the original map rather than reanalyzing each step. I may have to find the paper.
Now that I think about it more, since after targeted shrinking to remove an area, they say they expand it to keep the same size. Since other areas expand, it must be basing it off after shrinking rather than the original. I guess you don't want to do this too many times...
At first, I thought this was just simple image resizing. I was thinking "Someone should tell them that css can take percentage values for height: and width:"
That shows promise. I thought the deletion bit was pretty slick.
Copyright ©2000-2008 Jeremy Mooney (jeremy-at-qux-dot-net)
I am curious if it remembers the lines it has inserted or removed. If you enlarge and shrink and enlarge and shrink several times, and then go back to the original size, will it deteriorate or will it be the original image?