Re: [Lcms-user] Color Temp

[Andrej Javoršek]

Hello!

http://www.brucelindbloom.com/


All the necessary equations and theory is on that site.


Regards
Andrej Javorsek

On 2006.1.19, at 5:44, Alex Lau wrote:

I found very useful information about RGB to XYZ and further XYZ1 to XYZ2 transformations on mr. Lindbloom's site. Proper transform is probably RGB(input) to XYZ(input) Bradford transform to diferent color temperature XYZ(output) and transform to RGB(output). But be careful Color temperature is normaly in Kelvin scale, white point is most often as chromatic coordinates (small) xyz (x+y+z=1) but white point of illuminant can be also in CIE standardized values as (capitall) XYZ, that are in range from 0 to more than 100.

Thanks you so much for the reply

Marti wrote:

Hi,

Alex, I don't fully undestand all the math you are using here.
I think it would be better to separate things in order to make
it clear. Assuming you want to change the image white point
the steps would be:

Yeah .. I don't really understand 100% either, but yes I'm basiclly
alter the white point. (Colortemp?) So the white reference of
different white point can be calculated by using colortemp to XY
and XY to XYZ (assume Y=1). Is that ok?

a) Convert image to XYZ space. This can be done in several
ways. If the image is in sRGB then you could use the
gamma/matrix method, but not all images are in sRGB space.

Right coverting image to xyz for me is also a difficult choice
there are so many difficult matrix I can use but I pick the
Bradford...( I forget why already :) I remember something something
related to 2 and 10 degree observation point
So let say I get a digitial capture jpg output what should I
do to convert that RGB to XYZ?

b) If white points are close enough, a simple scale would
suffice. Somethink like:

  XYZout = XYZin * (WPout/WPin)

I think this is what I'm trying to do in step 4

if white points are not so close, then you would need
something more complex, like an appearance model.

is there any link I can understand more about that?

c) Convert back the XYZ to the desired output space,
which can be sRGB or any other.

right here I get to use the Reverse Bradford ... since I come
in XYZ with Bradford, with my logical guess I just assume what
I use to come, after transform I go back out RGB the same way :)

Bradford matrix is supposed to handle chromatic adaptation,
which is how our brain gets rid of white point. This is used
on perceptual & rel.colorimetric intents. I  think you should
use abs.colorimetric instead.

hum ...any example of abs colorimetric? I can try it on my calculation
and see it provide better result :)

I see F-Spot using lcms doing the exact same thing but the result is

Thanks

a little different ( not big just a little ) and wonder how you did it.

Regards
Marti Maria
The littleCMS project
http://www.littlecms.com



----- Original Message ----- From: "Alex Lau" <[EMAIL PROTECTED]>
To: <lcms-user@lists.sourceforge.net>
Sent: Tuesday, January 17, 2006 9:26 AM
Subject: [Lcms-user] Color Temp

Dear all
I'm new to the list.

I have wrote some color temp transform function.
And realized lcms able to do the same, I'm just trying clear my
misunderstanding and make sure my implementation is correct.
( even though it look very close to photoshop raw plugin already )

What I do is this

1) first I get colortemp from and to and make a xy ( assume no lower
then 2000k )
       if( colortemp >= 2000 && colortemp <= 7000 ){
               xyy->x =-4.607 * pow(10, 9)/ pow(colortemp, 3) +
2.967800 * pow(10,6) / pow(colortemp, 2) +
99.11 / colortemp + 0.244063;
       } else {
               xyy->x = -2.0064 * pow(10,9)/ pow(colortemp, 3) +
1.9018 * pow(10,6) / pow(colortemp, 2) +
247.48 / colortemp + 0.237040;
       }
       xyy->y = -3 * pow( xyy->x, 2 ) + 2.87 * xyy->x - 0.275;
2) Then XY to XYZ by assuming y = 1
       new_xyz->X = orig_xyy->x / orig_xyy->y;
       new_xyz->Y = 1;
       new_xyz->Z = (1-orig_xyy->x - orig_xyy->y)/ orig_xyy->y;
3) Then using BradFord matrix to get the white reference rgb
R = new_xyz->X * BradFord[0][0] +
    new_xyz->Y * BradFord[1][0] +
    new_xyz->Z * BradFord[2][0];
G = new_xyz->X * BradFord[0][1] +
    new_xyz->Y * BradFord[1][1] +
    new_xyz->Z * BradFord[2][1];
B = new_xyz->X * BradFord[0][2] +
    new_xyz->Y * BradFord[1][2] +
    new_xyz->Z * BradFord[2][2];

Reference for making the transformation matrix

matrix[0][0] = BradFord[0][0]*destRGB->R/srcRGB->R;
       matrix[0][1] = BradFord[0][1]*destRGB->G/srcRGB->G;
       matrix[0][2] = BradFord[0][2]*destRGB->B/srcRGB->B;

       matrix[1][0] = BradFord[1][0]*destRGB->R/srcRGB->R;
       matrix[1][1] = BradFord[1][1]*destRGB->G/srcRGB->G;
       matrix[1][2] = BradFord[1][2]*destRGB->B/srcRGB->B;

       matrix[2][0] = BradFord[2][0]*destRGB->R/srcRGB->R;
       matrix[2][1] = BradFord[2][1]*destRGB->G/srcRGB->G;
       matrix[2][2] = BradFord[2][2]*destRGB->B/srcRGB->B;

5) transformation matrix = matrix * Reverse_BradFord

Now we can start working on RGB

6) RGB->sRGB (by gamma apply)
if( RGB <= 0.04045 )
RGB / 12.92
else
pow ( ( (0.055 + RGB)/ 1.055),  2.4 ) ;

7) sRGB->XYZ (by multiple D65 transform matrix)
X = R * D65[0][0] + G * D65[1][0] + B * D65[2][0];
       Y = R * D65[0][1] + G * D65[1][1] + B * D65[2][1];
       Z = R * D65[0][2] + G * D65[1][2] + B * D65[2][2];
Here is the question.. I see you are using D50 transform matrix


8) XYZ * transformation matrix

9) XYZ back to sRGB

10) sRGB back to RGB

So basically is

RGB->sRGB->XYZ->Transform->XYZ->sRGB->RGB


or wrong :)

Thanks

Alex





-------------------------------------------------------

files
for problems?  Stop!  Download the new AJAX search engine that makes

4) New we get Source Color RGB white reference and Dest Color RGB White which is for printing color space? I'm not sure how to choose, any idea? I'm not sure what I'm talking about ... could someone tell me I'm right This SF.net email is sponsored by: Splunk Inc. Do you grep through log searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel? cmd=lnk&kid=103432&bid=230486&dat=121642

_______________________________________________
Lcms-user mailing list
Lcms-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/lcms-user

-------------------------------------------------------

for problems?  Stop!  Download the new AJAX search engine that makes

This SF.net email is sponsored by: Splunk Inc. Do you grep through log files searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel? cmd=lnk&kid=103432&bid=230486&dat=121642

_______________________________________________
Lcms-user mailing list
Lcms-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/lcms-user

smime.p7s
Description: S/MIME cryptographic signature