I'm starting to investigate the UniWB technic and I think I have found a method that allows to improve the UniWB.
As you may know the goal of this technic is to achieve a custom white balance that will not modify the raw values, in particular the red and blue values, so as the RGB histogram that will be displayed by the camera will be as close as possible to the raw histogram.
This white balance is usually refered to by 4 values RGGB (Red, Green, Green, Blue) and these values are generally x 1024 1024 y (or x 1 1 y if the values are normalised) and the objective is to achieve a white balance that provide 1024 1024 1024 1024.
I have tried the lens cap method and the method of blowing all the pixels but it doesn't work on my classic 5D.
So I downloaded a file from this site : http://www.guillermoluijk.com/tutorial/uniwb/index_en.htm
that was prepared using the magenta chart method described there. There are also lot of other files for other cameras.
Using this file I obtained the following values : 1015 1024 1024 986
This was good but I was wondering if it is possible to achieve a better result ? And the answer is yes ! After using the method I will describe below I was able to achieve the following values : 1016 1024 1024 1027. using the same downloaded file.
So here is the idea. It's very simple. It consists in using the White Balance SHIFT feature of the camera. This feature allows to correct the white balance.
So after setting the white balance to CUSTOM using the downloaded file, I did 8 shifts of the WB corresponding to the 8 surrounding points around the center of the axis that are displayed in the camera when you call the SHIFT/BRACKETING feature (use SHIFT not BRACKETING).
For each shift position, I took a shot and then using exiftool I wrote the obtained values. This gave me the following table :
997:991 1005:968 1016:944
1006:1009 + 1025:961
1016:1027 1024:1003 1034:978
After this first table, I saw that the results are better in the bottom left corner. So I decided to go 2 positions further from the center in that direction and took a serie of other shots and obtained a more complete table :
987:1012 997:991 1005:968 1016:944
997:1030 1006:1009 + 1025:961
1006:1049 1016:1027 1024:1003 1034:978
1016:1068 1025:1045 1033:1022 1043:996
The best position is the position that offers the smallest distance from the 1024 value in both channels. So in my case it is the position corresponding to 1016:1027.
Try it and let me know if it worked for you.
