Page 1 of 2

Sepia tunes?

Posted: Sun Oct 04, 2015 3:25 pm
by Zepper
According to this page, sepia tunes should use the following formula.

Code: Select all

outputRed = (inputRed * .393) + (inputGreen *.769) + (inputBlue * .189)
outputGreen = (inputRed * .349) + (inputGreen *.686) + (inputBlue * .168)
outputBlue = (inputRed * .272) + (inputGreen *.534) + (inputBlue * .131)
But instead of sepia, I'm getting... this. :( Why???

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 3:41 pm
by lidnariq
That matrix of parameters isn't singular. Looks like rounding error, though.

Code: Select all

>> [.393 .769 .189; .349 .686 .168; .272 .534 .131]^-1
ans =

   1.2727e+03   1.5455e+03  -3.8182e+03
  -1.9008e+02   6.1983e+02  -5.2066e+02
  -1.8678e+03  -5.7355e+03   1.0058e+04
Is there any chance you typoed or transposed one of the numbers?

You probably want to convert to greyscale and then use a simple colormap to convert that to sepiatone (White down through a grayish color, then through brown, then black)

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 4:21 pm
by tepples
If the effect you're aiming for is less "real is brown" and more "the past is brown", this should approximate sepia for an NES picture:
  • Map 20 and 30-3C to 37
  • Map 10 and 21-2C to 27
  • Map 00 and 11-1C to 17
  • Map xD-xF and 01-0C to 07

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 4:30 pm
by rainwarrior
These specific values are the values for sepia tone that are recommended by Microsoft.
What an odd statement. Is Microsoft was some kind of sepia authority now?

If you want zero blue in your sepia conversion, you should create your own matrix that completely desaturates the image to remove input colour influences on the output. The one you're using only partially desaturates.

I think lidnariq suggested what you really want. Convert to greyscale first, then some sort of filter/map to convert to brownish colours. (You might be able to do with with a single matrix, but you'll probably want to work it out in steps first, at least.)

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 5:32 pm
by Zepper
Fine, so I'm using my own current method for monochromes. :)

Code: Select all

newValue_R = (brightness * $70) >> 7
newValue_G = (brightness * $42) >> 7
newValue_B = (brightness * $14) >> 7
The sepia hex triplet (from wikipedia) is $70$42$14.
Result below.

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 6:07 pm
by lidnariq
Which looks more like an amber CRT than actual sepiatone, because sepiatone (a silver sulfide gelatin print) isn't a single line in RGB space. It's formed by producing a thicker and thicker layer of Ag₂S, which for a given amount W of material, blocks X% of blue light, Y% of green light, and Z% of red light. So it's almost a straight line in HSL (but not in HSV)

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 6:57 pm
by Zepper
lidnariq wrote:Which looks more like an amber CRT than actual sepiatone, because sepiatone (a silver sulfide gelatin print) isn't a single line in RGB space. It's formed by producing a thicker and thicker layer of Ag₂S, which for a given amount W of material, blocks X% of blue light, Y% of green light, and Z% of red light. So it's almost a straight line in HSL (but not in HSV)
But that's not "converting to grayscale firstly". How am I supposed to do with RGB values?
From this alternative page I've found "43.9% red, 25.9% green and 7.8% blue". With a "small kick of x2", I got the following result below.

Re: Sepia tunes?

Posted: Sun Oct 04, 2015 7:16 pm
by tepples
Try this nonlinear formula and show us how it looks:

Y = .3R+.6G+.1B
Gnew = Y
Bnew = (Y * Y / Ymax + Y) / 2
Rnew = 3 * Ymax - (Gnew + Bnew)

where Ymax is 1.0 or 255 or whatever for your particular color space

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 8:06 am
by Zepper

Code: Select all

   int Y = (p->red * .3) + (p->green * .6) + (p->blue *.1);
   int Gnew = Y;
   int Bnew = (Y * Y / 255 + Y) / 2;
   int Rnew = 3 * 255 - (Gnew + Bnew);
   p->red = Rnew; p->green = Gnew; p->blue = Bnew; 
   if(p->red > 255) p->red = 255;
   if(p->green > 255) p->green = 255;
   if(p->blue > 255) p->blue = 255;
Reddish. :(
Is something wrong?

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 8:21 am
by tepples
I typed that off the top of my head without testing, as I was on a rare break from crunch time on my present project.

Try newR = Y, newG = Y, and newB as before.
Or try newR = Y, newB as before, and newG = (newR + newB) / 2.
Or try newR = Y/2 + 128, newG = Y/2 + 64, newB = Y/2.

Or how does it look with my mapping everything to column 7?

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 10:54 am
by lidnariq
All of those won't have the desired effect because they are straight lines in RGB.

Convert the input to grayscale, use the grayscale volume as L in HSL (with H≈orange, S≈50%), convert back to RGB.

Another thought: Grayscale, then try using different gammas for the different channels. Perhaps R=2.2, G=1.1, B=.55? Produces this:
rgamma_2.2_ggamma_1.1_bgamma_0.55.png
rgamma_2.2_ggamma_1.1_bgamma_0.55.png (1.13 KiB) Viewed 4334 times
Anything with Rgamma = 2Ggamma = 4Bgamma looks not unreasonable, upon briefly playing with it.

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 1:39 pm
by Zepper
My method of grayscale takes the brightness level and uses it to generate the RGB values for sepia.
I don't know... that red * 2.2 seems to generate a couple of $FFs.
Anyway, result below.

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 2:06 pm
by rainwarrior
Gamma means a power/exponent operation. In this case I think the suggestion was:

R ^ (1 / 2.2)
G ^ (1 / 1.1)
B ^ (1 / 0.55)

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 2:32 pm
by Zepper
rainwarrior wrote:Gamma means a power/exponent operation. In this case I think the suggestion was:

R ^ (1 / 2.2)
G ^ (1 / 1.1)
B ^ (1 / 0.55)
Sorry for the request, but mind you to write a C code for it? Mine's generating grayscale only.

Re: Sepia tunes?

Posted: Mon Oct 05, 2015 2:35 pm
by rainwarrior

Code: Select all

float r; // input colours in the range 0-1

#include <math.h>
r = pow(r, 1.0 / 2.2); // R ^ (1 / 2.2)

// output r is still in the range 0-1