GIMP Version: 2.8.14
Operating System: Windows
GIMP Experience: New User

---

Hi all,

from time to time I make use of Gimp Color Value Invert native filter (duplicate layer, value-invert, strong gaussian blur on the inverted values layer, mode set to overlay, opacity to the appropriate percentage; this improves -when the case- the colours of an image)
Now my questions:
1-the inverted colours layer looks to me always "pixelated", at least in some areas, while the source looks not; is this due to a choice of the filter author to accelerate the operation? or is there any other reason?
2-is there an equivalent function in G'MIC? I'm quite sure there is, but I'm not able to find it...

Thanks!

I think you need to show us an example of this "pixelation" that you are seeing.

And also, does the layer you are colour-inverting have any transparency on it?

Kevin

OK
just taken one
"pixelation" seems to occur on the dark (original) values

original


inverted values


no transparency (as far as I see)

another example
original:


value inverted


still in the (originally) dark areas: maybe the original is also pixelated but my eyes do not see it?

maybe this behaviour on dark pixels is known around.
by googling I found a comment on a gimp script (old) which says:

GIMP: YCC Inversion by Stratadrake
Resources & Stock Images / Application Resources / GIMP Scripts & Plugins
The inversion technique I use to create all my inversions of traditional media is a three-step process neither difficult nor (very) time-consuming, but even so; here's a GIMP script to perform those steps automatically, with just a single click.
...

Overview
The script performs a brightness-inversion of the current image layer, swapping light and dark tones while leaving color data (hue and saturation) more or less unchanged. The end result is similar to the GIMP's "Value Invert" command, but without the color distortion produced at low RGB values

No, the pixellation IS present in the original photo ..if you zoom to 800% you can clearly see 8x8 pixel blocks ..it's probably just an artefact caused by the JPEG compression.

The Value invert just shows up the pixellation present in the photo much more ..compare with Invert ..you can see the pixellation after Invert at 800% but at 100% the edges merge ..after Value invert the edges don't merge!

I wanted to double-check the last comment I posted, concerning an old script
downloaded and installed
it works w/o changes even if old
the result (as the colours are concerned) is much different from standard value invert, but the effect to improve the colours of an image by overlaying this with the original seems good
maybe I had the opportunity to discover (and share) and old useful script

outcome from luma-invert

hence? is there a problem in the filter itself? does it mean that "value-invert" operates on an image at 800% then takes one pixel out of the 8?
I'm not a graphic expert, but following what you said, jontait, it looks so.
(Any image at 800% very likely is pixelated, you don't think so?)
also: what's your comment on the comment of the author of the script just found?

You are assuming that Gimp is working in "linear light", and AFAIK this isn't true (at least before 2.9, where "linear light" becomes an option).



What this means is that the same perceived amount of change at either end of the Value range doesn't correspond to the same number of units, so you end up with very few distinct values where your eyes require a lot more for a smooth appearance. Since "Invert" is the equivalent of a downwards diagonal in Curves, it could be replaced by an inverted gamma curve.

There is no fault with the filter as far as I can see, it's doing it's job correctly on the input image.

The problem is the JPEG "chunking" in the input image.
The JPEG compression limits the color variation in "chunks" without limiting luminosity dynamic range .. so any processing based on luminosity (such as Invert or luma invert) won't generally show up the inherent "chunking", but processing based on color values may well do.

This is a screenshot of part of the input image at 800% with curves adjusted:

As you can see, the image is clearly divided into "chunks" of 8x8 pixels ..that's just the way the JPEG compression works.
The Value invert just makes the chunking more obvious.

thanks Ofnuts for the video and for the news concerning the next release of Gimp
so far, as a simple user, probably I will switch to the luma invert filter for enhancing my photos

thanks jontait for the detailed explanation
do you consider my (possible) decision to use for the future the luma invert instead of the value invert a good one or a bad one?

Looks okay to me, Dinasset. And it probably is better for low RGB values than the standard Value invert (as the author commented).

thanks!

The following is from the PDB entry for plug-in-vinvert.
Note the sentence: Sometimes produces unpleasant color artifacts on images from lossy sources (ie. JPEG).

I use to use Value Invert a lot, but I now prefer Luminosity Invert.

http://registry.gimp.org/node/8352

Hmm..
That's actually the same script that Dinasset found, but with a different name and slightly different format!

At http://stratadrake.deviantart.com/art/G ... n-86515731
Filename: GIMP_YCC_Inversion_by_Stratadrake-1.scm

At http://registry.gimp.org/node/8352
Filename: invert-YUV.scm

Yeah; in the Color Drop down menu, it's called Luma Invert.

just playing with the luma_invert possibilities
it seems that the plug-in "recompose" does not work when in the "decompose" phase it has been selected one of the two YCC options suffixed 256
"YCbCr_ITU_R709_256"
"YCbCr_ITU_R470_256"
while it works when in the "decompose" phase it has been selected one of the other two YCC options without the suffis 256, id est:
"YCbCr_ITU_R709"
"YCbCr_ITU_R470"

maybe jontait is able to explain this to me (or there is really a problem/limitation in the "recompose" plug-in

(I was just trying to extend the script for luma-invert to all 4 possibilitioes)