Half a croissant, on a plate, with a sign in front of it saying '50c'
h a l f b a k e r y
A hive of inactivity

idea: add, search, annotate, link, view, overview, recent, by name, random

meta: news, help, about, links, report a problem

account: browse anonymously, or get an account and write.



Magic Image Enlarger

Enlarge small images and fill in the details using computer rendering
(+3, -3)
  [vote for,

Like in the movies, where they are able to magically enlarge or zoom in on a small image, even if the detail didn't exist. We'll use photo-realistic computer rendering to fill in the details.

This might be pretty hard to do: the way I was thinking of doing it is to create a 3D model from the image, identify textures, and render an identical scene in a photo realistic way, but generating details based on the type of image. Then we'd need a way to combine the rendered pixels with the original pixels.

I posted this here because it's too big of a project for me to do, and I think it's worth thinking about! :)

GeoffPeters, Aug 01 2008

photosynth http://en.wikipedia.org/wiki/Photosynth
[xaviergisz, Aug 01 2008]

sift - scale-invariant feature detector http://web.engr.ore...eature%20Detector]]
Could be used to build something like this. [jutta, Aug 02 2008]

Finding Paths through the World's Photos http://phototour.cs...n.edu/findingpaths/
(Thanks, nix!) [jutta, Aug 16 2008]

Multiscale Texture Synthesis http://www.cs.columbia.edu/cg/mts/
It's not going to be accurate, but... [mylodon, Aug 17 2008]


       Maybe it could search the Internet for similar images and extrapolate.
phundug, Aug 01 2008

       Magic indeed. Interpolation will only get you so far.
phoenix, Aug 01 2008

       I think if you could develop software that could look at a picture and recognize a 3D image, people would send you truck loads of money. The stuff after that would be just gravy.
MisterQED, Aug 01 2008

       The information provided by a 2D image is simply inadequate to construct 3D models from. Even people blind in one eye use multi positional interpolations to understand 3D objects.
WcW, Aug 02 2008

       Yes and no. A 2D image + an understanding of the world is quite sufficient to build a likely hypothesis. (For example, most people can look at a photograph and get an idea of what 3D-scene it depicts.) In our case, for "understanding of the world" read "large database of images".   

       I really love this idea - I wish someone would do this.
jutta, Aug 02 2008

       Such a concept works really well for architectural images where conventions like flat surfaces and right angles are common. Construction of a 3d model of a room from a single picture is actually rather easy using a tool in photoshop called Vanishing Point. For a computer to interpret more complicated images a vast index of 3D models is required. But how large is that lawn gnome? Is it a tiny one held in front of the camera or a full size one implanted in the ground? Forced perspective questions arise for all objects that do not fall on surfaces that have a relationship to the horizon.
WcW, Aug 02 2008


       You are using 'photoreal computer rendering' as a magic technology and not providing any real ideas.   

       In fact, you include 'Magic' in the name -- so blatant, so shameless!   

       It's like proposing a new fuel efficient car that will be 'really efficient' because of 'solar power and advanced technologies'... try again.
mylodon, Aug 03 2008

       I guess this is more of a concept than an idea (and an old one at that), and bordering on WTCTTISITMWIBNIIWR, and if one ingores the magic technologies mentioned, it's practically plain old WIBNI.
Spacecoyote, Aug 03 2008

       I had a similar thought along these lines concerning fractal image compression, which incidentally i've probably misunderstood. Anyway, if a fractal can be found which describes, say, a tree, you could zoom in on one of its leaves to reveal something which resembles the real vein structure on one of its leaves. Moreover, suppose an object can be described by two different, interacting fractals such as a blood vessels in nerve tissue or saliva in a duct. You could then zoom in on both, providing apparently new information in the interaction.   

       I think this could work but would probably supply spurious information much of the time.   

       On another note, i think that if you'd simply avoided using the word "magic", people wouldn't have focussed on that, though they might have ended up creating it as a spurious detail.
nineteenthly, Aug 04 2008

       Wait... Are you making things zoom in really well, or are you making 3-D models, or both?   

       If you're just hoping to zoom in like in the movies, the technology has been around since the Viking era. It's known as Starburst Pixel interleaving technique. What they do is, you scan all the pixels in the picture, then stretch the image so that each pixel is surrounded by nine blank pixels. After that, the computer calculates the difference between the originally adjacent pixels, and colors the new pixel an average tone between the two. It's very good at adding detail that isn't there, and has been used for nearly 30 years now to fuel the Cydonia face, and pyramid city myth.
ye_river_xiv, Aug 17 2008

       //This might be pretty hard to do://   

       Yep, more significantly it is very hard from one image. There is a certain ambiguity in translating any image into a 3D shape. Think of those optical illusions (or Escher works). It is, however, not impossible. You just might find it more cost effective to use another approach. Like two or more pictures (photosynth), or interleaving and interpolation techniques.
4whom, Aug 20 2008

       I think taking a succession of pictures, each one zoomed closer to the subject then the last, would work excellently.
mylodon, Aug 21 2008


back: main index

business  computer  culture  fashion  food  halfbakery  home  other  product  public  science  sport  vehicle