Resize Image Contract For Free

[Laughter] seam carving is a technique for dynamically resizing images not just for cropping them or for stretching them disproportionately but to resize them in a way that's aware of the content inside them so we have this full-size image and then if we we can resize it down vertically here and we haven't just chopped that bottom chunk of the image off we've actually pulled out intervening elements of the image where it's not visually noticeable and the result is that this image you're seeing here even though it's almost half the height looks visually coherent and it doesn't look cropped it just looks kind of like the mountains got closer together and the clouds kind of all lined up next to each other for this new Cinema project with ibeam and craters project we were exploring themes of flooding and water kind of intervening into spaces that someone's trying to cross and so it occurred to me that this technique was perfect because you could kind of expand and contract water as that distance needed to shrink or expand in all these examples I'm working in processing and I'm working with a library that I'd written in processing called dynamic programming which is optimization technique that has lots of applications one of which is seam carving kind of the coolest of which is seam carving caring so this sketch is really designed to illustrate how seam carving works we start with this original image on the right here this kind of strange painting which was one of the original images in the original paper and this Edge detection algorithm we ran in the beginning that's how we chose the important parts you can see this mask on the left the edges of it really show up as being valuable and the faces of both the people like those are the parts we've decided that we want to keep of the image that we want to distort last because edges of images are something that you notice expensive means that it's part of the image we really don't want to eliminate so the goal is to find the path from the bottom to the top that's the cheapest and we could do that in this incredibly iterative way where we try each possible one but there'd be zillions and zillions of them and that would take forever so that's why we generate this middle picture here it's basically the result of pre- doing all that stuff in this kind of efficient way with this dynamic programming technique so what this image really represents is at each pixel it represents the cost of the cheapest path from there to the top top so the best path the place that interferes with the least of the important parts of the image from any pixel to the top will be the darkest one but you see as we go through here I'll go through and I'll eliminate seams so that red line is showing which pixels we're eliminating and then you can see the image on the right just stays rectangular but it shrinks down because if we subtract one pixel from each row of the image image is still rectangular but just one pixel less wide so if I...