Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: Combining RGB Separations to Create Colour Image  (Read 4030 times)
RFPhotography
Guest
« on: October 09, 2012, 10:52:25 AM »
ReplyReply

I'm trying to play around with RGB separations to create a colour image.  Since I don't have any true 'Prokudin-Gorsky' type negatives I'm taking an RGB image and separating each channel.  This website has a bit of information on the process but I'm obviously missing something.

Starting with my RGB image, I've highlighted each channel and copied it into a new greyscale document.  Then I've created a new greyscale document, as the link above outlines, and pasted each of the separations in as layers with red on the top, green in the middle and blue on the bottom.  What I don't know is what comes next.  How does this get converted into an RGB image?

Thanks.
Logged
Stephen G
Jr. Member
**
Offline Offline

Posts: 94


« Reply #1 on: October 09, 2012, 12:11:20 PM »
ReplyReply

Once you've aligned the grayscale Red, Green and Blue layers (which you don't need to do, of course, in your example) then create a new RGB document and copy each of the grayscale layers over to the channels of the new RGB document.
Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #2 on: October 09, 2012, 12:25:28 PM »
ReplyReply

How funny, I did this exercise a while back after seeing this exhibit mentioned on the web.

1) Use the composite 3 panel B&W image, extract each panel of the three-image composite and paste each image onto its own layer in a RGB document.

2) make sure you figure out which panel corresponds to which color channel (RGB).

3) Select all three layers and use Edit > Auto-align layers... to register them.

4) Apply a gradient adjustment layer above each of the three B&W panels, and clip the gradient adjustment to the B&W under it.

5) For the appropriate channel, make the gradient adjustment layer's gradient go from black to pure color (for example, 0,0,0 to 255,0,0 for the red gradient adjustment layer).  If you get funky results, reverse the gradient (check the "reverse" tick box).  You can apply a different "filter" color by adjusting the RGB value of the end of the gradient.

6) Working form the bottom to the top layer, the base (bottom) layer blend mode NORMAL, 100% opacity.  Middle = NORMAL, 50%.  Top = NORMAL, 33%.
6a) Put a curves adjustment above everything to balance luminosity.

OR

7) Set the blend modes of the middle and top layers to LINEAR DODGE (ADD) and opacity 100.  No curves adjustment for luminosity necessary.

Have fun!

Kirk

« Last Edit: October 09, 2012, 01:20:25 PM by kirkt » Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #3 on: October 09, 2012, 12:32:34 PM »
ReplyReply

Here's what I came up with when I did this exercise previously, after cloning out color plate defects, etc. and balancing color and tone.



kirk
Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #4 on: October 09, 2012, 12:39:06 PM »
ReplyReply

Here is a link to hi-res scans of the original B&W image composite:

http://www.loc.gov/pictures/resource/prok.01887/

kirk
Logged
jeremypayne
Guest
« Reply #5 on: October 09, 2012, 01:11:22 PM »
ReplyReply

Here's an older discussion of the same topic ...

http://www.luminous-landscape.com/forum/index.php?topic=37705.0
Logged
RFPhotography
Guest
« Reply #6 on: October 09, 2012, 01:32:09 PM »
ReplyReply

Ah ha!  Pasting back into the respective channels in the new document makes much more sense. 
Logged
Kumar
Sr. Member
****
Offline Offline

Posts: 653


WWW
« Reply #7 on: October 10, 2012, 03:54:10 AM »
ReplyReply

There's a long discussion here

Kumar
Logged

kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #8 on: October 10, 2012, 08:50:39 AM »
ReplyReply

You can also replace the gradient adjustment layers in the above layer stack with Photo Filters, uncheck the "Preserve Luminosity" box, and crank the filter up to 100% density.  This gives you a little more of an intuitive approach in changing the RGB value of the filter through which the B&W is being processed and seems to give better color reproduction.

kirk
« Last Edit: October 10, 2012, 08:52:37 AM by kirkt » Logged
RFPhotography
Guest
« Reply #9 on: October 10, 2012, 08:53:01 AM »
ReplyReply

Pasting the respective separations back into the designated channel in an RGB document works just fine.  It's easy and simple.  No need to make the process unnecessarily complex.
Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #10 on: October 10, 2012, 09:41:56 AM »
ReplyReply

... unless the filters through which the B&W images were shot were not pure 255 Red, Green or Blue.  Then the color derived from the B&W pasted into the channels will be incorrect.

The above process gives you the flexibility to prescribe the color of the filter through which color information will be derived from each B&W so you do not introduce problems into the composite that you have to undo with further adjustments.  It also provides some insight into how the filtered B&W images are combined to get a full-color comp. 

The above process is not complex, but go with what works for you.  As with most things in Photoshop, there are several ways to accomplish the final result.  This topic has been discussed a few times here, but I figured I'd share my experience with the process because you asked.

Have fun!

kirk

Logged
RFPhotography
Guest
« Reply #11 on: October 10, 2012, 10:26:16 AM »
ReplyReply

Given what I'm experimenting with, as outlined in the original post, the possibility of a different coloured 'filter' is not possible.

That aside, I looked at your method and can't see how it could possibly work.  I tried it, in fact, and it doesn't seem to.  I wonder if you've written your methodology the way some old grandmothers write recipes and have left out an ingredient or two.  Grin

What I end up with is either a muddy cyan colour obscuring the actual image if the grad layers are left at 100% opacity or a muddy cyan cast over the image if the grad layers are changed to the 66/33 opacity levels as the green and red image layers.  But the image itself is not turned into a colour image. 

Are the grad layers from 100% opacity to 0% (black to colour) or 100% to 100% (black to colour).  Again, I've tried both and neither works.

Why a grad as opposed to a simple colour fill layer?

The Photo Filters makes more sense.  The grad doesn't make a lot of sense to me.  The Photo Filter actually does work somewhat.  A lot of additional work needs to be done to get a proper colour balance and luminance level (H/S, Curves, Levels, Color Balance).  But better than the grad which doesn't work at all.
Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #12 on: October 10, 2012, 11:17:37 AM »
ReplyReply

Hopefully this will give you a better recipe than "season to taste!"  Wink

First attachment is the three source panels from the single negative linked above in a previous post (Library of Congress).  The assumption is that red, green and blue filters were used for the separation, so, if we make some assumptions about the subject, we can infer which plate is which.  If we assume a caucasian-like skin tone, then blue will give us a dark tonal value for skin, green a mid tone and red a light tone.  This helps as a first pass guess for the recombination of the separate plates in the negative

Second attachment is a stack using Photo Filters.  The key to the adjustment layer approach is to make sure that each adjustment layer is CLIPPED to the B&W image below it so that the adjustment (gradient, photo filter, etc.) will only affect the layer directly below it and not the whole stack.  For anyone using PS that is not aware of how to clip the layer, make sure the adjustment layer is directly above the layer you want to operate on, hold down the Option (ALT on PC) key and hoever over the border between the two layers in the layers panel - the cursor will change to an "L" with a downward pointing arrow - click on the border and the adjustment layer will become indented in the stack, with the arrow to its left indicating that the layer is clipped to the one below it.

Photo filter settings: 100% density (or season to taste) and uncheck "preserve luminosity".  Filter layer blend mode: normal, 100% opacity.

Layer blend modes: bottom layer: normal, 100% opacity - middle and top layers: linear dodge (add), 100% opacity.

Give this a shot and see if it works.  I understand this is more complicated than necessary for the approach where you extract the channels from a color image using the B&W representation in the PS Channels panel.  In that case, all plates are registered and the channel color is exactly known.  However, if you want to composite Prokudin-Gorsky' type negatives, you may want a little more control over the compositing process.

Hope this helps!

kirk



« Last Edit: October 10, 2012, 11:19:31 AM by kirkt » Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #13 on: October 10, 2012, 11:25:21 AM »
ReplyReply

What I end up with is either a muddy cyan colour obscuring the actual image if the grad layers are left at 100% opacity or a muddy cyan cast over the image if the grad layers are changed to the 66/33 opacity levels as the green and red image layers.  But the image itself is not turned into a colour image.  

Are the grad layers from 100% opacity to 0% (black to colour) or 100% to 100% (black to colour).  Again, I've tried both and neither works.

Why a grad as opposed to a simple colour fill layer?

The opacity of the gradient layers should be 100%, normal blend mode.  The B&W image layers (in normal blend mode only) have the different opacity assigned (bottom=100, middle=50, top=33).  If you use linear dodge on the plate layers, all are 100% opacity.  Again, make sure you clip each gradient layer to the associated B&W plate.

The gradient approach is just a different method of mapping color to the tones in the B&W plates.

I agree, the photo filter method is better.

kirk
« Last Edit: October 10, 2012, 11:27:01 AM by kirkt » Logged
RFPhotography
Guest
« Reply #14 on: October 10, 2012, 12:26:32 PM »
ReplyReply

Yeah, I understand the opacity of the grad layers (or photo filter).  It was a question of the opacity of the grad itself.  When you create a grad you have colour stops and opacity stops.  You're setting the colour stops to black and red, green or blue.  It's the opacity at each end of the scale that's the question.  Standard is 100 to 0.  But that wouldn't really seem to have much of an effect.  It would depend on whether black was at the 100 opacity end or the colour was at the 100 opacity end.

The Photo Filter method works much better and when used with the Linear Dodge blend mode on the top two layers is pretty good.  The only thing I had to do after that was apply an H/S adjustment layer.  With the Normal blend mode and lowered layer opacity, much more work had to be done with additional adjustment layers - again, simpler is better. 

Still don't get the grad idea and can't get it to work.  The Photo Filter is a nice twist though.
Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #15 on: October 10, 2012, 02:17:11 PM »
ReplyReply

Both the begin and end stop of each gradient is 100% opacity.  Here is a capture of the blue B&W with the blue gradient adjustment layer, pictured in the screenshot, affecting the B&W plate - similar to what you would see in the blue channel in an RGB image, if you enable "Show channels in color" in Photoshop preferences.  You could add intermediate stops and change the uniformity of the gradient, increasing contrast in that color channel if you so desire.

The adjustment layer method just makes things non-destructive - if, like with the Prokudin-Gorsky images, I don't know exactly what they are supposed to look like in the final comp, I like being able to configure a few different plate mappings (filters, gradients, etc.) and toggle among them to see the effect of per-channel adjustments.

kirk
« Last Edit: October 10, 2012, 02:22:46 PM by kirkt » Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3580


« Reply #16 on: October 11, 2012, 03:56:35 AM »
ReplyReply

Both the begin and end stop of each gradient is 100% opacity.  Here is a capture of the blue B&W with the blue gradient adjustment layer, pictured in the screenshot, affecting the B&W plate - similar to what you would see in the blue channel in an RGB image, if you enable "Show channels in color" in Photoshop preferences.  You could add intermediate stops and change the uniformity of the gradient, increasing contrast in that color channel if you so desire.

Hi Kirk,

A small remark, if I may. Gradients with 'Smoothness' set to 100% will exhibit a bit of an s-curve to them. For a truely linear gradient, which may be partcularly useful in this case for increased control, Smoothness is best set to 0%.

It's easy to check that behavior when a 256 pixel horizontal gradient is made, going from black to white. Then the horizontal pixel position should correspond with the pixel value.

Cheers,
Bart
« Last Edit: October 11, 2012, 10:30:44 AM by BartvanderWolf » Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #17 on: October 11, 2012, 08:00:05 AM »
ReplyReply

Makes sense about the "S" curve nature of a smooth gradient - the gradient techniques tends to produce more punchy color, so this would explain it.  Thanks Bart!

Attached are two renderings, one with smoothness of all three gradients set to 100 and one with all three gradients set to 0.  The 100 smooth gradients image has more color contrast, as Bart's comment would predict.

kirk
« Last Edit: October 11, 2012, 08:05:54 AM by kirkt » Logged
RFPhotography
Guest
« Reply #18 on: October 11, 2012, 10:32:00 AM »
ReplyReply

Here's one using the Photo Filter adjustment layers.  The unaltered RAW file is included for comparison.  The combined version has a bit of a red cast but that's easily corrected.  Overall not bad.  Switching the blue and red filters to Deep Blue and Deep Red gives it a more neutral balance.
Logged
kirkt
Full Member
***
Offline Offline

Posts: 172


« Reply #19 on: October 11, 2012, 05:40:26 PM »
ReplyReply

Yeah - if you click on the color swatch for "blue" or "red" filters, etc. the RGB values are not pure 0,0,255 or 255,0,0.  Deep red is 242,0,0, for example.  You can specify a custom color if you want, and make it 255,0,0 for pure red, for example.  The way the black and white channel image is rendered will also affect the final composite as well (dot gain, LStar, etc.).

fun, eh?!

kirk
« Last Edit: October 11, 2012, 05:42:44 PM by kirkt » Logged
Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad