Many years ago, I ran across this site and was intriqued with this concept. I was using Sample Colorize but I would inevitably get way too many tones. Only recently (like an hour ago; lol) found out why (you will have to see the tut for the settings). Anyway, Click the link to the tut and you too will be Pseudo Greying (see tut for a cool test that you can try yourself as well and you too will be a believer).

Link to Tutorial: http://justpaste.it/pseudog_gimp

ref: http://www.fontplay.com/freephotos/thir ... 108-19.jpg

OK; manage to create a gradient with exactly 1786 tones. Count them yourself using Color Cube analysis. I was trying to do a diagonal gradient and never thought to just do a vertical.

I had a mind to write a script to perform this... and actually did so (code is in spoiler below), however, it would seem that the Palette Map plug-in does not properly handle palettes with more than 256 colors; so my script does not work. I am only including it as an example of how not to do it (the generation of the palette is correct, though, and should be more accurate than obtained through the gradient method).

I guess one would have to generate a temporary image with the pseudo-gray palette and then use Sample Colorize (I may do this in the future, but have other things occupying my time at the moment). Note that it was not too long ago that Sample Colorize could not effectively be called non-interactively so if script or plug-in is written using this approach, be aware that it will only work with GIMP 2.6.11 or later.

When you do a black and white gradient map (dragged) properly, it will be a Pseudogrey image as I showed in my followup link; that's what's cool. Once you have a good target (a pseudogrey gradient or the pyrimid that I actually used in the tut) it's child's play for Sample Colorize. I have since noticed that sometimes I have to still run Sample Colorize twice to get 1786 colors (not for the target in the tut though). Still, no real big deal.

Wish I could have the desire to learn how to program Saul, but lost my appetite since I got out of Engineering (and really since college; didn't really have to do much coding as an Analog EE). I'm an idea's guy anyway. I wonder if this could be done in G'MIC?

If you were to make a layer above your greyscale image fill it with white and then run the RGB noise filter with all the color sliders set all the way up to 1.0 and then reduce the opacity of the layer to 2.0 and merge down you can achieve even more shades as shown in the colorcube analysis. But while it looks greyscale it is RGB.

I find this colorcube and sample colorize to be very useful to me in creating smooth raster patterns for 3D height map images. What I do is all the steps above but I also blur the merged image by a setting of 5. Then I do the Sample Colorize as shown in the tutorial and it comes out even smoother.

It actually changes the numbe of colors again, (reducing it) but this seems to result in a smoother image than just doing the blur at a setting of 5.

Thanks for the info Lyle.

Alan

In blend Dialog there are two check-boxes, Dither and Adaptive Super-sampling.

If neither is checked you will get 256 colours.
If both are checked you will get between 1745 and 1754 depending of them supersampling sliders, possibly out of that range, even.
With just Dither Checked I've seen as little as 1660 and as much as 1786

-Rob A>

Good things to know Rob; never really played with those settings.

I have corrected an error in the script in my previous post. I originally had the green and magenta "near gray" values transposed (as exhibited in the following graphic, green has a higher luminosity than magenta.)

Just curious Saul; have you made any progress in getting your script to output true Pseudo-grey?

No, I haven't; however, it appears to be possible to create a gradient with thousands of segments. So perhaps it would be feasible to modify my script to create a pseudo-gray gradient (with 1785 segments) which can then be applied using 'plug-in-gradmap', with the caveat being whether gradmap retains more than 8-bit precision when it calculates luminosity.

Thanks for the followup Saul. Might be a minor inconvenience, but using Sample Colorize interactive doesn't take much effort and it's quick too. Still, it requires a good target image; if the target has any posterization or noise, then there's little benifit to convert that result into a Pseudogrey anyway. Still, I'm quite pleased at some results that I've already shared (other forums) using the flow as I outlined in the tut.

Well, I'm a little late to the game, but you've managed to suck away 2 hours of my morning...



This will just define a black to white linear gradient, create a layer, apply the gradient across it, then sample colorize between the gradient layer and the active layer, with requisite options.

I've checked quickly and the indeed the result is 1700+ levels.

Thanks for this topic, it's been quite interesting!

Magic number is 1786 patdavid; just tried out your Script-fu and got the magic 1786 result. Thanks a heap; will update my tut thread with your Script-fu link which, I hope you don't mind, saved as a .scm file that's attached.

Excellent!
Don't mind at all - that's what I attached it here for. I think I may go put it up on the plugin registry as well.

Definitely good news patdavid. I'm a big advocate for pseudogrey (and, though not here, got hit for being over zealous; lol).

I've gone ahead and done a quickie writeup on your method (with attribution and links):

True PseudoGrey in GIMP

Going to PM you some additional things Pat and again, thanks for the Script-fu.

...You reminded me something with this grey mode attempts, but beware, my question is related to color profiles so there's always a chance that what I see on my screen does not match what you see, because of discrepancies between monitors, browsers, embed color profiles in png and all that jazz that usually annoys me when I think about it.

Briefly, I want to know if your pseudo grey method aims at converting hues to grey while _keeping_ the value of colors, because in your output greyed image, I can see that colors are translated to grey but without matching the value of the original image (eg: the trunks I see became darker in the grey output).

And my question is not purely theoretical, if someone created a script that can achieve a grey translation without value loss, I want it badly !

It does use Hue transistions to achieve the 1786 tones anarkhya. Unfortunatly, 8-bit images can only have 256 values if you will so when you monotone an image, you only will get 256 unique values. What Pseudogrey does is blend the transitions between values giving you a cleaner transistion. As a test, create a 1000X1000 image and run the gradient tool (default forground/background setting that is black to white) diagonally then count the number of unique tones using color cube analysis and if done right, you will have 1786 unique tones despite created a black to white gradient (run Max colors and you will see the colors show). That's what convinced me that Pseudogrey is the way to convert b/ws. The only way to get a higher number of values is to increase bit of image to 16 bits for example. One other thing though; if you save the result as a jpeg (lossy) you will loose tones; you need to save in a lossless format such as PNG to benifit from pseudogrey.

I'm a bit confused by...

Yes but.. Why talking about indexed image ? The photo shown is 32bit (or 24, I'm not sure, jpg is capable of 32 or 24 ?). As desaturating does not imply that you index the image (well it may do if you use mode grey, that's what you meant?) with the desaturate filter you obtain a non-8bit image.


Are you sure? I've used max quality jpgs (like 99% quality and all quality checkboxes ticked) a few times and, while it's technically still lossy, my eyes seem to be totally fooled.