pouët.net

ASCII representing gray scale

category: gfx [glöplog]
What's the best character sequence to represent the gray scale?

I mean... If you know any char sequence, it is fine. :P
added on the 2011-04-14 22:53:50 by Danguafer Danguafer
That depends on the actual charset/size/char spacing/font. See this prod for the best sequence you can get on a ZX 81 ;-)
added on the 2011-04-14 23:00:07 by JAC! JAC!
Danguafer: could you be a littlebit more specific?
anyway if i understand what you mean, you should try out different ramps. there's no exact answer to this i believe. one picture would look better with one ramp, and another picture would look better with another one. also there are ways to extract your best character combinations by calculating their average and combine, this is the easy way to find your characters. but since pictures are different and shapes are different some differencies need to be taken. but you can make your easy tool in a short time.
added on the 2011-04-14 23:04:30 by rudi rudi
.|..
added on the 2011-04-14 23:06:29 by havoc havoc
Rearrange and order the font by the number of filled/unfilled pixels.-. Mega grayscale!
added on the 2011-04-14 23:14:43 by linde linde
or make your own dither fonts if your platform is dos. just find a way to manipulate the bios fonts. :P
added on the 2011-04-14 23:23:08 by rudi rudi
I'm not sure if I understand the question either.
added on the 2011-04-14 23:34:27 by Gargaj Gargaj
Being more specific...

I was looking for a gray scale character ramp to print into the Windaz' console. Found this little one:

" .-:=+*#@". It's OK for me.

Thanks. :-)
added on the 2011-04-15 01:21:26 by Danguafer Danguafer
"@MBHENR#KWXDFPQASUZbdehx*8Gm&04LOVYkpq5Tagns69owz$CIu23Jcfry%1v7l+it[] {}?j|()=~!-/<>\"^_';,:`. "

That's based on the amount of filled pixels with.. Arial, size 8, I think.
added on the 2011-04-15 01:27:07 by msqrt msqrt
Console? Text mode? mmhh. Lucida... and...

There's tons of combinations. You could even use dbqp and colors to get kind of a curvature.
added on the 2011-04-15 01:56:26 by yumeji yumeji
░▒▓█
Danguafer: isn't that font-specific though?

Then again, Sol might have a better answer if he's around.
added on the 2011-04-15 02:29:13 by Gargaj Gargaj
It's tricky. As several have mentioned, it's font specific, but it's not just "how many pixels are on or off". Figuring out the actual brightness of a character is more art than science, or if there's some science in there, I haven't figured it out. That said, all my algorithms just try to play with pixel counts, which leads to one set of solutions, while some other folk have made hand-tuned grayscale ramps (likely by some non-coder ^_^) and have resulted in more eye-pleasing results, at least with "simple" color ramps (i.e, not full rgb color space conversion).

As an example, think of a 8x8 block with two lit pixels. If the pixels are at the center and next to each other, it'll look brighter than if the two pixels are in opposite corners. Of course, if you fill the whole screen with these, you'll end up with pixels touching again, which is (yet again) different brightness. Hm, maybe I should blur the whole screen at once, to let glyphs bleed to their neighbors.. food for thought, eh? =)

My braindumpish research can be read here: http://iki.fi/sol/textfx/. The main differences between mine and radman's approaches are that he uses the whole charset while I like the ascii subset more, and he scales down while I blur; which approach is better is a matter of taste. I'd welcome similar writeups from other folk who have done rgb->ascii conversion libs.

And now, to blow this thread up totally and to make these lines hard to read, here's one fun graph I did at some point, just trying to figure out the x/y pixel balance of characters.. what's to take away from the graph is that the charset isn't all too balanced, but that's about it.

BB Image
added on the 2011-04-15 08:02:16 by sol_hsa sol_hsa
It's font specific, but I think there are some generic sets like " .0#". :z

Sol: My good man, I noticed that you are very studied in the ways of text art. Thank you for explaining so much details about it all.
added on the 2011-04-15 09:58:17 by Danguafer Danguafer
@Danguafer: well, I didn't do a thesis on it, but I recall someone else doing just that.. =)

And all of these are just tinkering with things, I'm pretty sure someone with computer vision / codec background could do better.
added on the 2011-04-15 15:35:45 by sol_hsa sol_hsa
Radman?!
added on the 2011-04-16 05:08:15 by trixter trixter
sol_hsa: have you tried figuring out character brightness by comparing the amount of contiguous black area vs contiguous white? Seems to me that an imaginary character that is a white square filling half of the character cell would be calculated as 50% grey when you average the pixels, but it appears much brighter than a checkerboard pattern to the eye.

Perhaps your algorithms would benefit by preferring noisier characters, a simple loop though the character pixels counting the flips between black and white would offer a good measure of how noisy a char is.
added on the 2011-04-16 17:15:48 by Claw Claw
Here's the result of my playing:

BB Image
added on the 2011-04-16 21:13:34 by Danguafer Danguafer
I was just trying to experiment Sphere Tracing with Text Art.
added on the 2011-04-16 21:14:27 by Danguafer Danguafer
@trixter: or maybe we just should let some artist figure out some cool grayscale ramp =) Or, find someone with codec/computer vision background to give some actual algorithm for this..
added on the 2011-04-16 21:26:43 by sol_hsa sol_hsa
I think you can get some pretty good results if you combine the count-pixels method with some shape analysis on the image, i.e. a weighted color level vs adherence to the original image.
added on the 2011-04-17 21:43:31 by linde linde
maybe even more so if you add some blur to the char with a low-pass filter before comparing to the image
added on the 2011-04-17 21:45:11 by linde linde
Hahaha, guess who just started reading sol_hsa's excellent article! :D
added on the 2011-04-17 21:53:55 by linde linde
Depends on the font. Calculating weighted coverage works generally well.

BB Image
added on the 2011-04-18 02:13:25 by cgr cgr
nice and pure monochrome. but now add some greyscale ramp colors and do some eigenvector shit on the chars bits to make it perfect.
added on the 2011-04-18 02:38:31 by yumeji yumeji

login