Slashdot Mirror
Pixel Inventor Goes Back To the Drawing Board
lawpoop writes "Russell Kirsch, inventor of the square pixel, goes back to the drawing board. In the 1950s, he was part of a team that developed the square pixel. '"Squares was the logical thing to do," Kirsch says. "Of course, the logical thing was not the only possibility but we used squares. It was something very foolish that everyone in the world has been suffering from ever since.' Now retired and living in Portland, Oregon, Kirsch recently set out to make amends. Inspired by the mosaic builders of antiquity who constructed scenes of stunning detail with bits of tile, Kirsch has written a program that turns the chunky, clunky squares of a digital image into a smoother picture made of variably shaped pixels.'"
This sounds like the ongoing debate between analog and digital audio. Everyone likes using images like these during the debate, but given enough resolution (bits), the closer the digital audio will be to its original analogue (electrical) source.
I'm god, but it's a bit of a drag really...
Actually, you jest, but I remember the first time I saw footage from WWII that was in colour and being stunned, because it was so vivid.
And, then there was the Russian guy who created colour photos in 1909 using techniques he created himself.
There are more things in heaven and earth, Horatio, Than are dreamt of in your philosophy.
Lost at C:>. Found at C.
1) A pixel isn't "invented" by anyone. A pixel is just a concept that is so straightforward, like the wheel, language and adding numbers. It's not a question of which single person "invented" it. It's just a question of, once the technology is there, it WILL be used, no matter what.
2) What kind of screen are you going to use for that? Each pixel can have different types of pixel sizes so no screen could fit that. A square grid is the most uniform division of 2D space into units.
3) If this would have been about hexagonal pixels, I'd have found this cool.
4) At best, this is a new compression scheme for storing pictures - but certainly not a way to display them (see 2))
5) Non square pixels are not a new idea, see for example sensors of cameras.
blatant as it may be, I read the article three times now - and Soilworker, you did well not to bother. I'm pretty sure the answer is not in there.
This doesn't seem to be about square pixels in terms of display technology (where hexagonal pixels may indeed be superior).
It also doesn't seem to be about picture acquisition.
On the face of it, it seems to be talking about mapping rudimentary shapes to pixels so that they conform to a most-likely contrast-matching scenario with regard to surrounding pixels. Which some other posters here already pointed out with posts about JPEG and the like - but it's not really comparable to that either. Not in technique and not in performance.
At best, as far as I can take away from it, it could be a different way to display an image when zoomed in / a technique that could be used when enlarging an image to provide greater apparent detail (although you wouldn't want to enlarge it - you'd want to store the masks found with the original image for display).
The results in the news blurb look pretty decent and if nothing else 'different' from other 'smart scaling' methods, so it's worth exploring. But what this has to do with square pixels as we're mostly familiar with them, I have no idea.
Now, about those hexagonal display pixels...
First, here's the actual paper, since it clarifies what exactly he's suggesting and doesn't seem to be linked anywhere in the article.
It's not a suggestion that we start using non-square pixels for displays or cameras or scanners or what not, though he's certainly not being very clear about anything and the reporting on this is just making matters worse. What the paper proposes is a method where:
1) The image is split into 6x6 blocks
2) For each block, you go over the four rotations of the two following two-section masks:
The triangular mask:
ABBBBB
AABBBB
AAABBB
AAAABB
AAAAAB
AAAAAA
The rectangular(ish) mask:
BBBBBB
BBBBBB
BBBAAA
AAAAAA
AAAAAA
AAAAAA
for a total of eight effective masks, and average the values under each section, resulting in two values, A and B.
3) For the mask and rotation that has the largest difference between A and B, you output the mask, the rotation, and the A and B values, resulting in 19 bits from a 6x6 (288 bits) block.
Though he talks of non-square pixels and whatnot, it's really just a compression algorithm. A really stupid one. Basically it's a bad variation of vector quantization, with lots of baffling details. Why 6x6 blocks? Why those specific masks? Why are you maximizing contrast instead of minimizing error like any sane person would do, WHY? There's no rationale given for any of these choices, not theoretical, not empirical, not even subjective.
The same sort of rigor extends to his comparison, where he compares his compression algorithm to, instead of, say, another compression algorithm, the image apparently simply downscaled and then scaled back up. And not even with a halfway decent resampling algorithm, but with nearest neighbour. Not to mention that the "non-square pixels" version has 2.375 times as many bits to work with. If he'd done a comparison to a reasonably modern compression algorithm like JPEG, the results would be much less favorable to him.
tl;dr Some old guy put together his My First Compression Algorithm kit and it's being treated like a revolution in graphics by ignorant reporters. Nothing to see here, move along.
Pixel was completely misused in the article. He's working an image scaling algorithm for photos. That isn't saying that it's not noteworthy, interesting or important; it looks like it works great and I'm not aware of anything that produces results that good on photos. There is the Hqx family of filters, but those were designed for emulators and aren't meant to be used with more than 256 colors.
Actually, anti-aliasing is nothing like blurring. True anti-aliasing is actually a projection of a higher sample rate to a lower one by combining more than one sample within the area of a single sample at the lower sample rate. While not as accurate as the higher resolution image, it is significantly more accurate than simply selecting one sample from each area. Blurring would be taking selecting one sample within the area of a single sample at the lower rate, and then averaging neighboring samples, which means you actually end up with less information than the un-blurred un-anti-aliased image.