Talk:Error diffusion

Gamma
On all documents I have found so far, the dithered images look lighter or brighter than the originals. Watch that at an more or less 50% grey area in the original image! This is because practically all digital images are coded to visible brightness. 128 in an 8bit scale means 50% visible brightens. Normally the computer screen makes the conversion (gamma). With black/white dithering the screen has only black and white pixels. So the visible image has not the necessary computer screen gamma anymore. To correct that you must apply the computer screen gamma (for example according to sRGB) to the digital image before you dither it. The missing gamma in the dithered image adds to the typical appeal of these image, but if you really like to reduce the difference between the original image and the dithered one, you need to add gamma! May be someone can check that and add a note in the article. Michael Tuchscherer --134.3.156.38 (talk) 03:32, 31 December 2012 (UTC)

Sharpening
The discussion on edge sharpening seems it could benefit from an examle including a continuous tone original, an error diffused version, and one halftoned using traditional clustered dot. Any takers? Lovibond (talk) 22:56, 2 June 2008 (UTC)

Error "added to" or "subtracted from"?
The article says that the error in quantizing one pixel is "added to" the next pixel along before it is in turn quantized. Shouldn't this be "subtracted from"?

For instance, say we're dealing with black (0) and white (255) allowed output levels, and we have a sea of light gray (192). The first pixel will be quantized to 255, with an error of 63. We want at least some of the following pixels to be quantized to 0, but if we add the error we'll get an even larger value, with no chance at all of being quantized low. If we subtract the error, however, then the pixels will be


 * 1) 192 => 255 (error 63): white dot
 * 2) 192 - 63 = 129 => 255 (error 126): white dot
 * 3) 192 - 126 = 66 => 0 (error -66): black dot
 * 4) 192 - (-66) = 258 => 255 (error 63): white dot

and you'll get approximately one black dot in four, which is correct. -- Dan Griscom (talk) 04:11, 29 October 2008 (UTC)


 * It is ambiguous, as the error can be defined as either the unquantized value minus the quantized value (what you wanted, minus what you got), or as the quantized value minus the unquantized value (what you got, minus what you wanted). If you would like to close this ambiguity, it would be a better article. Perhaps adopting the same convention as one of the early papers would be a way to approach it. Lovibond (talk) 14:31, 31 October 2008 (UTC)

No, "added to" is correct. The error values added to the neighboring pixels can be positive or negative.RGB213 (talk) 20:05, 18 June 2019 (UTC)

Untitled
fyi. http://omohundro.files.wordpress.com/2009/03/omohundro90_floyd_steinberg_dithering.pdf

the pseudo code is :

val<-get_pixel(x.y)+error_arr[x+1]

if val>128

then set_output_pixel(x.y); error<-val-255; --output white

else error<-val;

omit — Preceding unsigned comment added by 61.219.197.130 (talk) 10:05, 18 October 2012 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified one external link on Error diffusion. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20131022220241/http://michal.is/projects/image-dithering-in-matlab/ to http://michal.is/projects/image-dithering-in-matlab/

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

Cheers.— InternetArchiveBot  (Report bug) 08:05, 23 September 2017 (UTC)