One of the things I studied in grad school was how to find the least visible seam in an image. This would be useful say if you overlapped two images and wanted to find the best way to cut them together, or if you wanted to create your own waldo image. I thought that the seam finding algorithms were really cool, but that there wasn’t a whole lot of use for them outside of doctoring photos and generating texture. Holy crap was I wrong. Check out this video:
This is the first time I’ve ever watched a technical video and thought to myself “I understand everything that is going on here… why didn’t I think of this!!” That’s a truly awesome idea. It makes me want to breakout my seam finding code and make an image resizer (I’ve written code that finds the low energy image seams you see in the video – it’s what I used to make the waldo image, which is the same image randomly pasted over and over again with a seam between overlapping areas).
The only problem with that idea is that finding seams isn’t as fast as they show you, at least not for large images (400+ by 400+). They might be using a different algorithm then what I’m thinking of, or maybe they were running the app on a super fast computer, but in my experience finding a seam can take some time if the image is of a decent size (by “some time” I mean 1 or 2 seconds). Though I might be wrong and this might be a super fast algorithm. It’ll be interesting to see how people apply this.
Have you worked with Image Magic at all?
The library? A couple of people recommended it to me when I started out my graphics projects, but I didn’t give it much of a chance. I think I tried to compile with it once and it didn’t work so I was like “screw this, I’ll just write my own image class”. Which was a little more fun for me, but it limited me to using ppm files (it would have taken too long to learn a more advanced image format).
If ImageMagik gave you access to the raw RGB values for each pixel in the image then it’d be an awesome library. I have no idea if it does this or not, but when working with algorithms like the one above, that’s basically the key thing you need.
I think the above technique would make a GREAT feature for photoshop or gimp. Sometimes you read about new technology, and it seems really cool, but then it also seems a little impractical unless you’re a huge organization. Like this new paper on scene completion:
http://graphics.cs.cmu.edu/projects/scene-completion/
That also falls into the category of mind blowing, but you need a database of millions of images for it to work (and probably a decent amount of time). The content-aware resizing seems like it could easily be made available to the public.
Yeah I use to hit up your site all the time back in the day. Then I stopped and forgot about it (no offense) because you didn’t do to much with it for a few years. Then today I visited your domain and was like whoa is this the same dude. I read your about page and yup. So I don’t know what the point of this was but whatever. Have fun.
haha, yeah, lately I’ve been having someone message me about once every 2 weeks saying they just stumbled upon the site again. I can’t believe I let it die, but I think I just painted myself into a corner, and college sort of took over, but oh well. Hopefully I can rebuild it into something better.
yeah i still remember after that one redesign you did away with the midi player everyone was like NNNNNNNNNOOOOOOOOOOOO