Reading mode: Light Dark

The Science of Color-cast fixing (need tips)

Go to last post
A

Arjun Roychowdhury

Senior Member
Messages
1,950
Reaction score
1
Location
US
Greetings, Please see:
http://www.pbase.com/image/23644172

This is an example of a picture I had taken with an obvious color cast problem. Till now, Ive been very unscientific with my processes - I kept working on color casts till my eyes said they were ok.

I would like to now understand how to do the same 'scientifically'.

Based on a tutorial I read:

One way to do it is to examine the Info dialog to see the R/G/B values of a known white spot or dark spot and try to make sure the R/G/B are similiar in numbers.

In this example, I sampled the white flowers in the lower left corner (below the ladies) as well as the white window in the background.
Both samples showed an equal mix of R/G/B values.

In addition, I also sampled the black pillar to the left, that too showed a correct R/G/B mix (7/7/7)

Therefore, I should assume this picture has no color-cast ?
This does not seem right. Im obviously missing something.

Can someone please set me straight ?
regds
arjun

--
http://www.pbase.com/arjunrc
 


First your image had a lot of noise so I filtered it with NI. There was an orange cast which was taken care of doing a color adjustment with curves. However, that left the image some what flat so I adjusted the midtones on color range using green as my starting poing and adding to the red and subtracting the blue. This brought some of the cast back but I have a feeling it is closer to the acutal colors you see in the building considering the stone the walls are made from and the lighting. Anyway, it gave more depth and dynamic range to the image. I then flattened and sharpened it. Personally I have found that most color casts are taken care of completely or at least to a great degree just by doing a color correction in curves. Do not use auto, do it the ole fashion way using threshold to find your black and white points and then hunt for your 128 gray point(that can take some time and doesn't always work).
 
Greetings Semo,
First, thanks for taking a shot. It sure looks good.
I am trying to nail down a process to have lesser dependance on my eyes
and hence have specific questions for you. Hope you would not not mind
answering them.
SEMO said:
There was an orange cast which was taken care of doing a color
adjustment with curves. However, that left the image some what
Click to expand...
This is my first problem. I also would have opened up my curves box, switched to the Red layer and pulled down my curve to reduce reds.

However, when I 'sampled' my black and white points, they had an equal mix (almost) of R/G/B. So why then would I assume there is an orange cast ? (I know it does, Im just trying to understand why my process of numbers is flawed).

One of the things I tried was this:

a) Option 1: I let my 'eyes' select a black and white point by setting color samplers in the original image. As I mentioned, the points I selected did not show a color variance.

b) Option 2: I used the threshold dialog to select my whites and blacks. And indeed, using this methods, the black and white spots I selected had a color variance. When I then adjusted the RGBs of those sampled images, a lot of the red cast was removed.

I therefore conclude that for selecting reliable white/black pts, threshold is the way to go, and not randomly clicking what 'looks' white or black in the original image.
Is this correct ?
SEMO said:
doing a color correction in curves. Do not use auto, do it the ole
fashion way using threshold to find your black and white points and
then hunt for your 128 gray point(that can take some time and
doesn't always work).
Click to expand...
Once I had set the colors correctly on any one point (be it black or white) why do I need to search for other points ? Is it not right that fixing a color cast on any one sampled point should uniformly correct overall color cast, since I assume color cast is not selective when I take a picture ?

regds
arjun

--
--
http://www.pbase.com/arjunrc
 
Arjun

You can eyeball black and white areas in your image but your RBG must read 0,0,0 for black and 255,255,255 for white in your info pallete. Thresholds just makes it easier and more acurate. But with these two points you have only adjusted you highlights and shadows. You need to include your midtones. Often you will see your most dramatic shifts in color when you click on your 128 point. Sometimes it will be too much which usually means you didn't nail the point on the head so zoom in on it. If you can't find a 128 point you have two choices(well, at least that's all I have). Adjust you midtones in curves eyeballing it, but here we get back to the original problem of what looks good to who. The other is look at your histogram in each of the RGB channels. Write down the numbers. Generally green falls in the middle with red higher and blue lower. Go to color range layer and add the difference between green and red in the red channel and subtract the difference in the blue channel. Once again you may still want to adjust the opacity or blending mode but I rarely find this neccessary.
 
Arjun Roychowdhury said:
Greetings, Please see:
Click to expand...

Arjun Roychowdhury said:
This is an example of a picture I had taken with an obvious color
cast problem. Till now, Ive been very unscientific with my
processes - I kept working on color casts till my eyes said they
were ok.

I would like to now understand how to do the same 'scientifically'.

Based on a tutorial I read:

One way to do it is to examine the Info dialog to see the R/G/B
values of a known white spot or dark spot and try to make sure the
R/G/B are similiar in numbers.

In this example, I sampled the white flowers in the lower left
corner (below the ladies) as well as the white window in the
background.
Both samples showed an equal mix of R/G/B values.

In addition, I also sampled the black pillar to the left, that too
showed a correct R/G/B mix (7/7/7)

Therefore, I should assume this picture has no color-cast ?
This does not seem right. Im obviously missing something.

Can someone please set me straight ?
regds
arjun

--
http://www.pbase.com/arjunrc
Click to expand...
Probably it is picking the appropriate sample pixels. If the image is noisy you could pick just the noise. Here are the 4 pixels I measured with the color sampler tool:

The RGB values are:
  1. 1 83/64/66
  2. 2 117/109/107
  3. 3 239/234/230
  4. 4 245/239/239


It looks that GB values are fairly equal, only the red values are higher and need to be lowered. Then I made curves adjustment layer, selected the red channel and changed the values all at once 83/65, 117/108, 239/232, 245/239. Then I selected RGB channel and lightened it a little 185/226.



I hope this helps

Regards
 
Arjun Roychowdhury said:
http://www.pbase.com/image/23644172
Click to expand...
[...]
Arjun Roychowdhury said:
Therefore, I should assume this picture has no color-cast ?
This does not seem right. Im obviously missing something.

Can someone please set me straight ?
Click to expand...
What you are seeing as a "color cast" is the fact that the foreground of this picture is illuminated by daylight, while the background is lighted by incandescent lights.

The points you chose to test are both in the foreground, so the colors were what you expected. But the background is a lot yellower.

There are various ways to deal with this. In my opinion, the easiest is:

Get a color sample of something you want to make more or less neutral. Use a 5x5 average. I chose the light vertical pillar on the left.

Make a new layer. Fill it with this color. Invert it (Image-> adjust-> invert). Now, in this case, you should have a light blue. Put the layer in "hue" blend mode. Adjust the opacity slider until the background is more neutral in color.

Add a layer mask to this layer, "reveal all". Paint over the foreground with a black brush, to bring it back to its original colors.

That's it. It's quicker to do than to describe.

--Al Evans
 
Hi
Following is your original (for ease of reference)



Following is my correction by using the gray eye-dropper on a select single pixel



as per tutorial from this website-link= http://www.dpreview.com/learn/Image_Techniques/Color_Correction_Speed_01.htm
Hope this helps
Cheers
Ash
Arjun Roychowdhury said:
Greetings, Please see:
http://www.pbase.com/image/23644172

This is an example of a picture I had taken with an obvious color
cast problem. Till now, Ive been very unscientific with my
processes - I kept working on color casts till my eyes said they
were ok.

I would like to now understand how to do the same 'scientifically'.

Based on a tutorial I read:

One way to do it is to examine the Info dialog to see the R/G/B
values of a known white spot or dark spot and try to make sure the
R/G/B are similiar in numbers.

In this example, I sampled the white flowers in the lower left
corner (below the ladies) as well as the white window in the
background.
Both samples showed an equal mix of R/G/B values.

In addition, I also sampled the black pillar to the left, that too
showed a correct R/G/B mix (7/7/7)

Therefore, I should assume this picture has no color-cast ?
This does not seem right. Im obviously missing something.

Can someone please set me straight ?
regds
arjun

--
http://www.pbase.com/arjunrc
Click to expand...
 
PS

I haven't bothered with any noise correction, sharpening etc. since your question was mainly about color/cast correction
 
The difficulty with this image is that you have 4 different light sources with different color temps that you are trying to balance:

(1) Your flash - slightly bluer than daylight
(2) The daylight coming in through the windows
(3) The overhead incadecents - your orange

(4) The flouresents (offsets sitting on top of the two doors in the background) - add some green

This makes it impossible to do a single correction that takes care of every part of the image. You either have to decide which part you want balanced to your liking and leave the others alone, or you have to do some masking and adjust the different areas separately.
Arjun Roychowdhury said:
Greetings, Please see:
http://www.pbase.com/image/23644172

This is an example of a picture I had taken with an obvious color
cast problem. Till now, Ive been very unscientific with my
processes - I kept working on color casts till my eyes said they
were ok.

I would like to now understand how to do the same 'scientifically'.

Based on a tutorial I read:

One way to do it is to examine the Info dialog to see the R/G/B
values of a known white spot or dark spot and try to make sure the
R/G/B are similiar in numbers.

In this example, I sampled the white flowers in the lower left
corner (below the ladies) as well as the white window in the
background.
Both samples showed an equal mix of R/G/B values.

In addition, I also sampled the black pillar to the left, that too
showed a correct R/G/B mix (7/7/7)

Therefore, I should assume this picture has no color-cast ?
This does not seem right. Im obviously missing something.

Can someone please set me straight ?
regds
arjun

--
http://www.pbase.com/arjunrc
Click to expand...
 
You must log in or register to reply here.

Keyboard shortcuts

f
Forum
m
My threads
Mobile site
About
Editorial content
Cameras & Lenses
Community
All content, design, and layout are Copyright © 1998–2025 Digital Photography Review All Rights Reserved.
Reproduction in whole or part in any form or medium without specific written permission is prohibited.
When you use DPReview links to buy products, the site may earn a commission.
©GPS Media - Guides, Products, Services.
Back
Top