Between the round and the square
June 28, 2010 3:25 PM Subscribe
This comment on the article made me laugh:
Posted by: SimpleMind | 06/28/10 | 5:23 pm |
Title should be: “First man to scan an image re-invents image compression”.
posted by mrnutty at 3:40 PM on June 28, 2010
Posted by: SimpleMind | 06/28/10 | 5:23 pm |
Title should be: “First man to scan an image re-invents image compression”.
posted by mrnutty at 3:40 PM on June 28, 2010
That could be THX-1138's baby picture.
posted by longsleeves at 3:41 PM on June 28, 2010
posted by longsleeves at 3:41 PM on June 28, 2010
His new method looks a lot like the pixel art scaling algorithms that are built into a lot of emulators.
posted by zsazsa at 3:42 PM on June 28, 2010 [3 favorites]
posted by zsazsa at 3:42 PM on June 28, 2010 [3 favorites]
This article is very confusing. The description of the masking algorithm reminds me a bit of HQX, a way of scaling up pixel art in game emulators to 2x or 4x the resolution. It's basically a form of edge detection done in a discreet grid. It can guess wrong, but often makes pixel art look better.
posted by Nelson at 3:43 PM on June 28, 2010 [1 favorite]
posted by Nelson at 3:43 PM on June 28, 2010 [1 favorite]
Ooh, thanks for that link, zsazsa! Very neat.
The real message here is that we need a camera that detects only changes - a differential pixel. For more on that, check out Jack Tumblin's "Why I want a gradient camera" [PDF]
posted by fake at 3:56 PM on June 28, 2010
The real message here is that we need a camera that detects only changes - a differential pixel. For more on that, check out Jack Tumblin's "Why I want a gradient camera" [PDF]
posted by fake at 3:56 PM on June 28, 2010
A Pixel Is Not A Little Square, A Pixel Is Not A Little Square, ... (pdf)
posted by gruchall at 3:56 PM on June 28, 2010 [1 favorite]
posted by gruchall at 3:56 PM on June 28, 2010 [1 favorite]
It's like mpeg, only... less. An interesting fundamental departure from "square" pixels, happens when you displace every other row by one half pixel, making them effectively hexagonal. We're too far down the rectilinear grid for that now, but it hexagonal pixels have some advantages in quality for natural images, as well as some complications.
And pixel data isn't square at any rate, the captured samples are, or are supposed to be, Gaussian spots, the square pixels are a raster display artifact. If you imaging pixels are square, put your lens out of focus by one half pixel.
posted by StickyCarpet at 5:12 PM on June 28, 2010 [1 favorite]
And pixel data isn't square at any rate, the captured samples are, or are supposed to be, Gaussian spots, the square pixels are a raster display artifact. If you imaging pixels are square, put your lens out of focus by one half pixel.
posted by StickyCarpet at 5:12 PM on June 28, 2010 [1 favorite]
Read the article and all I could think was 'that blockhead.'
posted by JoeXIII007 at 5:44 PM on June 28, 2010 [1 favorite]
posted by JoeXIII007 at 5:44 PM on June 28, 2010 [1 favorite]
I hate to break it to him, but the world largely does not run on square pixels. We're talking here about a 2D sampling algorithm and the world does realise that the resolutions of the two need not be the same. I regularly work with anisotropically sampled data.
If we're only talking about images though, square pixels (assuming you don't want to go hexagonally) generally make sense because camera builders can't assume less resolution in one particular dimension.
Some food for thought: iPhone pixels (and a few other LCD displays) are not square (see the big thread yesterday about Retina Display); they sometimes have an aspect ratio that's close to 3:1! Plenty of VGA video modes have non-square pixels. Lots of printers have non-square pixels and scanners are particularly interesting, often having fixed resolution in one axis (the line-CCD) and variable resolution in the other (the stepping axis).
One more thing: pixels are not rectangles or squares, they are point-samples.
posted by polyglot at 5:51 PM on June 28, 2010 [2 favorites]
If we're only talking about images though, square pixels (assuming you don't want to go hexagonally) generally make sense because camera builders can't assume less resolution in one particular dimension.
Some food for thought: iPhone pixels (and a few other LCD displays) are not square (see the big thread yesterday about Retina Display); they sometimes have an aspect ratio that's close to 3:1! Plenty of VGA video modes have non-square pixels. Lots of printers have non-square pixels and scanners are particularly interesting, often having fixed resolution in one axis (the line-CCD) and variable resolution in the other (the stepping axis).
One more thing: pixels are not rectangles or squares, they are point-samples.
posted by polyglot at 5:51 PM on June 28, 2010 [2 favorites]
He called into Science Friday a few weeks ago.
I heard that. He called out of the blue. I thought "WTF, why doesn't Ira Playdough talk to HIM for a little bit?" But no....
posted by Floydd at 6:21 PM on June 28, 2010
I heard that. He called out of the blue. I thought "WTF, why doesn't Ira Playdough talk to HIM for a little bit?" But no....
posted by Floydd at 6:21 PM on June 28, 2010
He's also sorry for all that cancer. He was just on a roll.
posted by Astro Zombie at 8:02 PM on June 28, 2010 [2 favorites]
posted by Astro Zombie at 8:02 PM on June 28, 2010 [2 favorites]
Dude, don't apologize. Pixels are awesome, and inspired an incredible amount of culture.
posted by breath at 8:20 PM on June 28, 2010
posted by breath at 8:20 PM on June 28, 2010
I regularly work with anisotropically sampled data.
Regular 2D grids are always anisotropic. Resolution is a function of rotation.
I used to keep a gamma 1.0 colorsync profile around for debugging computer graphics stuff. Turn it on, and see if the moires and aliasing disappear. It made it really easy to see which programmers understood sampling and blending (answer: none).
posted by ryanrs at 10:14 PM on June 28, 2010 [1 favorite]
Regular 2D grids are always anisotropic. Resolution is a function of rotation.
I used to keep a gamma 1.0 colorsync profile around for debugging computer graphics stuff. Turn it on, and see if the moires and aliasing disappear. It made it really easy to see which programmers understood sampling and blending (answer: none).
posted by ryanrs at 10:14 PM on June 28, 2010 [1 favorite]
« Older World Wide Wes | Someone forgot to tell 'em the Cold War ended.... Newer »
This thread has been archived and is closed to new comments
posted by bhnyc at 3:35 PM on June 28, 2010