Yet another Siggraph video that blows my mind -- "Seam Carving for Content-Aware Image Resizing" by Shai Avidan and Ariel Shamir.
(via AE Portal News)
Part of what's so interesting is the idea of finding patterns in images/video that allow a system to be "content-aware". The use of edge detection to determine important vs. non important visual information reminds me of the process of analyzing time-sliced video footage to interpolate when cuts, dissolves, or wipes take place. Eddie Elliott has a great example of this on his website. Essentially, if you take all the frames of a video and place them back to back to make a cube, by looking at the sides of the cube you can see where a left to right wipe, a cut to different looking footage, or a cross dissolve take place. Someone told me a couple years ago about a company trying to refine the technique in order to create a software program that logged archived footage procedurally, rather than hiring someone to plod through hours of footage.
Image from Eddit Elliot's Video Streamer page. For more on time-slice/slitscan artworks, Golan Levin has a good archive here.