Page 1 of 1
[RESOLVED] -intensity ... -threshold IM 6.8.7.5 beta Q16
Posted: 2013-10-31T14:25:25-07:00
by fmw42
I was under the impression that -intensity rec709luminance should produce a result with gamma=1 no matter which of the allowed functions it precedes. This was fixed for -intensity rec709luminance -colorspace gray, but does not seem to be fixed for any of the other allowed functions such as -threshold.
Is this a bug or do I misunderstand how -intensity rec709luminance (or rec601luminance) should behave for functions other than -colorspace gray. See my my subsequent post below?
# create test image
Code: Select all
imb convert -size 128x128 -depth 16 gradient: grad.png
imb convert grad.png -intensity rec709luminance -colorspace gray -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: grad_irec709luminance_gray.png
0.311763 Gray Grayscale 1
Testing with -threshold results in an incorrect gamma=0.4545
imb convert grad.png -intensity rec709luminance -threshold 50% -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: grad_irec709luminance_t50.png
0.265624 sRGB Bilevel 0.45455
I suspect this is the same for all the allowed functions, though I have only tested with -threshold.
Re: possible bug -intensity ... -threshold IM 6.8.7.5 beta Q
Posted: 2013-10-31T14:28:18-07:00
by fmw42
Similar results using rose:
imb convert rose: -intensity rec709luminance -colorspace gray -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: rosde_irec709luminance_gray.png
0.196013 Gray Grayscale 1
imb convert rose: -intensity rec709luminance -threshold 50% -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: rose_irec709luminance_t50.png
0.0723603 sRGB Bilevel 0.454545
Re: possible bug -intensity ... -threshold IM 6.8.7.5 beta Q
Posted: 2013-10-31T15:52:13-07:00
by fmw42
I might be wrong about my assumptions above. Was it is intended that -intensity was simply a channel mixing function without regard to linearity/non-linearity (colorspace of the input). If so, then I assume one must first use -colorspace RGB to get the linear result, such as the following. If so, that seems a bit inconsistent with how it works with -colorspace gray
imb convert grad.png -colorspace RGB -intensity rec709luminance -threshold 50% -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: grad_irec709luminance_t50b.png
0.265624 RGB Bilevel 1
If it is required to use -colorspace RGB -intensity rec709luminance, then I do not see why we even need the luminance versions. One could just use
convert image -colorspace RGB -colorspace gray result
rather than
convert image -intensity rec709luminance -colorspace gray result
and
convert image -colorspace RGB -threshold 50% result
rather than
image -colorspace RGB -intensity rec709luminance -threshold 50% result
The bottom line: Was -intensity (and -grayscale) intended just to be channel mixing functions without regard to linear or nonlinear results? If so, then perhaps it was a mistake to have -colorspace gray sensitive to the linearity from luminance or luma while all the other functions are not. And if we change back to that channel mixing only pardigm, are there consequences or differences if we eliminate the luminance arguments to -intensity and simply precede it with -colorspace RGB in order to get the linear results.
How do we make -intensity work consistently so that it is clearly understood how it works?
Re: possible bug -intensity ... -threshold IM 6.8.7.5 beta Q
Posted: 2013-10-31T19:35:12-07:00
by fmw42
To partially answer my own question. These two are not the same. So we cannot do away with rec709luminance.
imb convert rose: -intensity rec709luminance -colorspace gray -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: rose_irec709luminance_gray.png
0.196013 Gray Grayscale 1
imb convert rose: -colorspace RGB -colorspace gray -format "%[fx:mean] %[colorspace] %[type] %[gamma]\n" -write info: rose_rgb_gray.png
0.394629 Gray Grayscale 0.454545
The latter converts first to linear (RGB), but then -colorspace gray converts back to non-linear (sRGB)
So the question becomes, should -intensity rec709luminance when used with other functions require -colorspace RGB to precede it or should it automatically force a conversion to linear (gamma=1) like it does for -colorspace gray.
Re: [RESOLVED] -intensity ... -threshold IM 6.8.7.5 beta Q16
Posted: 2013-11-01T09:57:48-07:00
by fmw42
After all these post, I think I have a better understanding of -intensity. So rather than rocking the boat and changing anything. I am OK with it as is.