Ariel Shamir and Shai Avidan have presented the coolest digital image effect I have ever seen. “Seam carving” allows an image to be resized non-uniformly, so you can change the height to width ratio in the image without cropping, but also without distorting important features in the image (such as faces).
If I understood the demonstration correctly, the algorithm detects horizontal and vertical paths which span the whole image and have the least gradient magnitude along the span. When resizing an image, the pixels along these seams are stretched or removed, leaving the rest of the image untouched. Areas of an image that absolutely need to be preserved can be manually excluded, and the seam generating function will negotiate paths around those pixels.
You need to watch the video to really see this in action, but the technique can also be used to remove whole portions of an image without perceptible artifacts. Essentially, you could cut an Ex out of your old vacation photos and the rest of the image would distort itself, unnoticeably, to fill in the gap.
Seam Carving for Content-Aware Image Resizing – [via] Video: Link PDF: Link
High-res copy of the above video @ Dr. Ariel Shamir’s site – Link
ADVERTISEMENT