Where/how to learn about digital image processing

Processing
21

That ratio seems to say, that “hue begins after white”?
That is, when the common part (white) is removed.

I’m not entirely sure what that means, but I think the answer is “yes” for (I stress) the simple arithmetical systems. If we add the same constant k to all three channels (k can be negative), the ratio becomes:

( (mid + k) - (min + k) )
-------------------------
( (max + k) - (min + k) )

  (mid + k - min - k)
= -------------------
  (max + k - min - k)

  (mid - min)
= -----------
  (max - min)

… so the ratio is unchanged, so the hue is unchanged.

Given the two conditions I mentioned above, the first condition tells us which of the six segments in the hue circle we are in. The boundaries of the segments are: red, yellow, green, cyan, blue, magenta and back to red. For example, if red is the maximum channel and green is the middle channel, the hue is somewhere in the segment between red and yellow.

Suppose RGB values (R,G,B) are in the range [0,1]. We can subtract the minimum value from all three channels without changing hue. So now the mimimum is zero. Then we can divide all channels by the maximum without changing hue, so the maximum is now at 1.0. In our example, the blue channel is zero, and the red channel is one, and green is somewhere between zero and one: the RGB values are (1, G’, 0).

The middle channel then gives the angular distance of this hue to one of the boundaries. If green is zero, the hue is red. If green is one, the hue is yellow. For other values of green, the hue is orange or reddish-yellow or yellowish-red or whatever.

22

This is an outstanding resource! I have a lot to read now. Thank you very much!

1 Like
23

@turboscrew A willingness to learn and ask questions is the beginning of a wonderful journey. :nerd_face: Since you have much reading to do, I won’t offer you new resources. I just want to caution that terminology can be mixed or used to mean different things in the literature. Be careful not to assume meaning because the author(s) may be using their words differently. There are also mistakes and laziness (as in the authors don’t actually know what they are talking about). This happens in research papers much more often than you would think! That is why it is better to discuss what you have learned with other people, e.g., on this forum, to verify whether you and the sources you have read are indeed correct.

4 Likes
24

That’s actually wrong. The only thing you are allowed to do without changing chromaticity (that is, the combination of hue and chroma), is multiplication. Adding or removing light (aka offsetting RGB) will hit the Abney effect.

Hue is a psychoperceptual component: 17-22-067 | CIE

Any angular component that you make appear from rewriting the cartesian RGB coordinates in some flavour of polar coordinates is only loosely related to hue. You can see what it yields to travel the HSV space at constant H angle: A perceptual color space for image processing. Notice how blue takes a leap into magenta.

It’s generally wrong to think that RGB is connected in any way to color. It’s actually misleading to call RGB spaces “color spaces”. You can write ugly numerical fits (like mine Engineering | Color saturation control for the 21th century) that try to connect tristimulus (RGB or CIE XYZ) to color correlates, but this ugly “data-science” more than actual colorimetry and it’s still inaccurate.

RGB spaces express light spectra in the shape of a 3D vector. That’s all. It’s generally wrong to start mixing RGB spaces with any concept of saturation, hue, chroma, brightness, lightness, etc. You need to heavily massage RGB spaces to start drawing some connection between RGB tristimulus and actual color attributes.

1 Like
25

That’s actually wrong.

It is actually correct, for (I stress again) the simple arithmetical systems.

The only thing you are allowed to do without changing chromaticity (that is, the combination of hue and chroma), is multiplication. Adding or removing light (aka offsetting RGB) will hit the Abney effect.

I have been discussing the effect on hue only, not chroma. The Abney effect is a perceptual effect. I have not been discussing perceived hue.

26

There is no non-perceived hue. That’s by definition a perceptual attribute of color.

No. Just because RGB code values are numbers doesn’t mean that you can do whatever you want discarding the reality of what is encoded. Math here is only the language, not the reality, not the thing being described. The only thing that matters is the reality being represented, the rest is mere technicality. You can’t offset RGB values and say “hue is preserved”.

27

There is no non-perceived hue. That’s by definition a perceptual attribute of color.

It is a definition, but not the only definition.

You can’t offset RGB values and say “hue is preserved”.

I can, and I did. For a definition of “hue”.

28

You are of course free to shit on 2 centuries of art theory and 1 century of color science and reuse words out of their context to make up alternative facts that hold true only in your unspecified framework where maths define reality.

But that’s not helping anybody.

29

I’m sorry if I didn’t make the context clear. I have been discussing “hue” mostly in the context of simple arithmetical systems: HSL, HSV, HSI, HCL etc.

30

Seems to me that’s exactly what @afre saw coming, a difference in definitions of the same term. I nearly replied about it too, but this is exactly what anyone learning about this subject is likely to encounter.

It’s possible to handle the ambiguity if you understand it from the perceptual layer up, so I recommend starting with the colour science. I started from the sRGB side and made lots of bad assumptions because of it!

1 Like
31

I did point that out in my first post on this thread.

1 Like
32

I think I mentioned, I’m not after good looking pictures, but rather documenting facts. And I’m interested in astronomy, not landscapes or portraits.
And, at this point, rather just in “signal processing”.

And I’m aware that RGB covers colours as roughly as 12th roots cover musical scales.
Also, I’ve never heard of vawelength of brown colour. :smiley:

Maybe it helps to specify, that the systems I’m dealing with are RGB-based, and I’m trying to learn to process RGB (and gray value) data.

1 Like
33

Well, it was kind of specified already.

34

Brown is essentially “dull, dark yellow”. So the predominant wavelength of brown is the same as that of yellow, but there is also a load of other wavelengths, spreading into the blue wavelengths.

1 Like
35

But you have to distinguish RGB encoding (each pixel is stored as a triplet of R, G and B values), and the colour space in which you do your editing (which can use R, G, B as basic axes, or another system, like Lab). But at the end of the day, you will have to translate the values from your colour space to RGB triplets, as that’s what your screen works with.
So don’t confuse the way your image is encoded with what happens “under the hood” while you edit.

And another problem for you might be that you are not even working with “real colours” (depending on the kind of data you get in your astronomy images). As for the “signal processing” part, I’m not sure a photography forum is the best place to discuss that, as the aim of the treatment might be completely different.

  • Signal processing: get the important information visible, no matter how the final image looks;
  • photography: get a good looking image, as close to what you want as possible
    (OK, that’s a bit extreme perhaps :stuck_out_tongue: )
1 Like
36

I wonder if this old topic has a couple ideas about books then:

37

As for the “signal processing” part, I’m not sure a photography forum is the best place to discuss that, as the aim of the treatment might be completely different.

This is, by far, the most informative place I have come across, when it comes to my interests. Like I mentioned, I checked the contents of quite some classic books, but they seem to miss about half of what I’m interested in.

38

Back to the topic of perceived colour, this might be an interesting overview: Did your primary school teacher lie to you about colour? | News | Victoria University of Wellington

1 Like
39

Roger Clark has a lot of great articles on astro photography:
https://clarkvision.com/articles/index.html

More about gear and technique than digital processing, but he certainly does touch on processing methods throughout.

1 Like
40

That Burger & Burge (2016) looks good, but maybe later.