Combining channels

[Panpan]

When looking at an individual primary channel in Photoshop, the image is lighter. Can somebody help me understand this? Something that is not perceptually neutral is happening behind the scene.

For example, a b&w image's RGB channels all look the same when I turn on each one individually, but as soon as I turn on more than one, the image gets darker.

This happens even in LAB mode. If I copy the Lightness channel to a layer in Luminosity mode, the image gets lighter. Isn't that channel supposed to contain all the non-color information?

Isn't this a problem? For instance it sometimes happens in restoration that only one channel is good. I would then like to promote that channel to a layer. How would I modify the layer's curve to approximate multi-channel grey tones.

[mistermonday]

Panpan, that is just your perception. Open a new blank document and fill it with R=255, G=128, B=0. Yes, that's a bright orange. When you look at the indiv channels they are pure white, mid gray, and black respectively. The fact is that the information in each pixel of the channel is one component that makes up the composite color of that pixel. The pixel beside it may have 3 totally different component values. When you view the channel you are viewing a map of pixels and the % of luminosity of the color of the channel you are viewing. Your view of that channel is not necessarily representative of the composite RGB image. So if you view the Green channel the bright and dark areas / pixels only tell you where there is a high or low content of green.
I hope I didn't confuse you more.
Regards, Murray

[Panpan]

Thank you for your help mistermonday.

I understand the general concepts of color space. I gave examples where I thought the individual channels would be representative of the composite grey values, i.e. any of the three channels for b&w images in RGB mode or the lightness channel for any color image in LAB mode.

That is, I expected them to be representative in those situations. They are not. I suspect it may have to do with when Photoshop applies gamma correction.

[secretagents]

Originally Posted by Panpan

e.If I copy the Lightness channel to a layer in Luminosity mode, the image gets lighter. Isn't that channel supposed to contain all the non-color information?

The L Lab channel is simply not equivalent to the Luminosity blend mode which is calculated in an HSL color space.

For example, a b&w image's RGB channels all look the same when I turn on each one individually, but as soon as I turn on more than one, the image gets darker.

I am not quite sure what you mean here. When I select an individual color channel of a BW pic (with Photoshop 7) I see an 8bit greyscale image and it displays the same as the 3 channels 24bit BW pic but if I select two channels I get a view colorized in either yellow, magenta or cyan with no apparent change in luminosity.

[TommyO]

All,
I think what Panpan is referring to is when any image is converted to grayscale, then converted again to RGB. At that point, all channels are identical, and they look alike on screen. If you view just one channel (Ctrl-1, or ctrl-2...), then toggle to the composite (Ctrl-~), the composite looks slightly darker.

I'm not coming up with a good answer at the moment.
Anyone else ?

[Panpan]

Originally Posted by secretagents

The L Lab channel is simply not equivalent to the Luminosity blend mode which is calculated in an HSL color space.

Fair enough. CS5 added two new layer blending modes, but we don't yet have a Lightness one. Also, isn't the L in HSL Lightness? I just find it strange in LAB mode that the composite channel is darker than the single channel that carries all the information.
Looking at two channels of a b&w image does overlay a color when in RGB mode. Photoshop does this even when the "Show channels in color" preference is unchecked. It does not behave that way in LAB mode.

Originally Posted by TommyO

I think what Panpan is referring to is when any image is converted to grayscale, then converted again to RGB. At that point, all channels are identical, and they look alike on screen. If you view just one channel (Ctrl-1, or ctrl-2...), then toggle to the composite (Ctrl-~), the composite looks slightly darker.

I find working in color in Photoshop confusing. For example, PS does not make it easy to convert a color image to B&W if the only thing you want to do is take away the colors. The defaults give a darker image.

Try this.
Open a color image in RGB mode. Add a default B&W adjustment layer in Luminosity blend mode. Satisfy yourself that it is darker by clicking it on and off.
Now turn off the B&W layer and add a Hue/Saturation layer in Luminosity mode. The image gets darker as you slide down the Saturation, but the shades are different from the B&W adjustment defaults. You can get the same results by using Desaturate on a layer copy of the background and putting it in Luminosity mode.

You can get a better idea of how perceptually non-linear HSL parameters are in RGB color space by playing with Hue/Saturation sliders in when it's in Hue, Saturation or Luminosity blending mode. The Hue slider in Luminosity mode is particularly surprising.

In LAB mode, the sliders are much more independent. Desaturating in LAB mode does yield a B&W image that is perceptually the same shades as the color image.
However switching image modes is not convenient. Also the B&W adjustment layer is not available in LAB. The Hue/Saturation layer has the same sliders in a drop-down menu that B&W has all in one window, but the scrubber adjusts saturation instead of luminosity.

In the end I solved my problem by creating two presets for RGB mode.
If all I want a faithful conversion to B&W, I use the Channel Mixer preset at 30 for red, 59 for green and 11 for blue.
If I want to also fine tune color shades, I use B&W preset at reds 30, yellows 89, greens 59, cyans 70, blues 11 and magenta 41 and start scrubbing.

By the way, these presets are close to Photoshop's formula to calculate luma (aka luminence in Adobe-speak).

[secretagents]

Well, after looking more closely I don't think the Luminance blend mode is calculated in an HSL color space but rather in the YCC color space. On the Photoshop install CD/DVD there are some channel mixer presets in a Goodies folder. One of those is the RGB to YCC preset. If you apply it and then use the Y (red) channel as a layer in Luminance mode on top of the original image, there is barely a difference with the original. Also converting the image to grayscale mode and then using that grayscale as a layer in Luminance mode on top of the original image yelds almost exactly the same. What you've been doing with the channel mixer is I believe quite the same as what Photoshop does when converting to grayscale mode, it does apply different coefficients to the RGB channels for perceptual reasons. But I think the fact all those methods yeld grayscale images which do not change (or barely) the image when set as layers in Luminance mode over the original mode does not mean they are truer than the Lab L channel. Theoretically the Lab L channel should the most perceptualy close to it if I am not mistaken.

I could not try the B/W adjustment as I use Photoshop 7 and it doesn't have this tool but I played a bit with the free plugin SmartCurve, which allows to use curves in various color spaces and could obtain a grayscale image nearly identical to the two above by desaturating the image in the LCH and YCbCr color spaces.

[TommyO]

Panpan,

Sorry to return so late. But, I thought I would offer an explanation in very basic language, so that it might be helpful to all readers. So, apologies if it sounds too basic.

When looking at an individual primary channel in Photoshop, the image is lighter. Can somebody help me understand this?

The channels were never intended to be "perceptually" accurate. They are basically masks, which try to represent color intensities as a grayscale pixel. You stare at a channel and focus in on one pixel which appears medium gray. But internally the application sees "Green, 176". That's all it cares about... it would prefer to tell you that, but "a picture represents a thousand words"... or in this case a few million pixel values. I suppose Adobe could have just as easily put a histogram there. So, try not to "humanize" it too much. It is simply intended to give us a feel for where the relative strengths of that color exist within the image, like a map.

Only the composite should be perceptually accurate. The composite has undergone all the underlying algorithms to make it perceptually accurate.

This is less deceiving to our eyes when looking at the channels of a true color image. The channels are often very very different.

However, once an image has been converted to grayscale (even if converted back to RGB), it starts to play with our mind a little bit. A "map" of color intensity looks almost like the grayscale composite. Well it should, almost. All three channels must mapping identical internal values to all their pixels. Otherwise, you will start to get some color in the composite. Since the composite will also be represented as grayscale, and there is no difference in the channels, the composite should look very similar to either channel. There is still a bit of math to create the composite and add the "perceptual" component. You're probably very close with the "gamma" correction comment, at least with grayscale images.

Isn't this a problem? For instance it sometimes happens in restoration that only one channel is good.

We all have learned over the years to simply expect there to be some work to do if we are going to use one alpha channel to overwrite to information in another alpha channel, i.e. using the infamous calculations or apply image.

I would then like to promote that channel to a layer. How would I modify the layer's curve to approximate multi-channel grey tones.

That's not what we normally do. We use calc's or apply image to "copy" the good channel data into the other channel(s). Only then do we mess with curves, channel mixer, HSL... often on each channel individually. It can take a lot of patience to get the color to return as is was originally intended.