New project: camera color profile calculation

[quatch]

Here are some test pictures for the SX10IS, using the new beta firmware, and the save as DNG options, taken under the stipulations at the top of thread (http://chdk.setepontos.com/index.php/topic,390.0.html).

http://www.fes.uwaterloo.ca/u/akasurak/CHDK_SX10IS

[ArtDen]

Here are some test pictures for the SX10IS, using the new beta firmware, and the save as DNG options

Dng4ps2 can't read dng files for creating color matrix. It reads only RAW files without any meta information

[zosX]

After doing some shooting this week and finally importing the results this morning, I am very happy with my results. I think that the raws look a lot better than the jpegs due to there being more dynamic range as well. If anything the results are somewhat warmer than the jpeg, which I am completely ok with because I feel that in comparison, the jpegs seem somewhat cold and lifeless colorwise. Look at these two shots. Happy with the results? The matrix I came up with is a definite improvement at least....

[ewavr]

I can't calculate color matrix for SX10, results are always poop (on stage 1).
Original (12-bit) RAW+JPEG here - http://ewavr.nm.ru/chdk/for_test/sx10_calib/
Because current version of dng4ps2 cannot handle such RAWs, I converted it to 10 bit (converter).
Any help will be greatly appreciated.
p.s. RAW size is 3720x2772.

[snc]

same trouble as ewavr over here... ixus 970 raw + jpeg. note that this is the untouched 12 bit raw still... help, please?

files here:
http://www.zshare.net/download/57436971b85bd82e/

[ewavr]

same trouble as ewavr over here...

Finally I did it avoiding some squares (brown and violet). Of course, result is not brilliant.

[snc]

mmm yeah, turns out that "works", but it really shouldnt be like that, should it...

[gam3ra]

How you guys shoot with 12bit RAW? My CHDK version shoots in 10bit. 

[snc]

that depends on the sensor, actually - chdk cant magically add some bits. the sx10, sd990, sd890, and sd790 are some examples for cams with a 12bit sensor.

[ewavr]

same trouble as ewavr over here... ixus 970 raw + jpeg. note that this is the untouched 12 bit raw still... help, please?
files here:
http://www.zshare.net/download/57436971b85bd82e/

6. Set focus to infinity

[snc]

yup, got it... a fellow 970 user took them, cause my cam was gone for a while... took new images with my cam, but same problem. anyway, the sx10 and ixus970 sensors seem to be extremely similar, if not identical.
surely some more experienced photographer can come up with "better" images from the 970, but i am not sure that it would help...

[ewavr]

yup, got it... a fellow 970 user took them, cause my cam was gone for a while... took new images with my cam, but same problem. anyway, the sx10 and ixus970 sensors seem to be extremely similar, if not identical.
surely some more experienced photographer can come up with "better" images from the 970, but i am not sure that it would help...

@scn:

Try also semi-official color matix from G9 (http://lclevy.free.fr/cr2/):

#define CAM_COLORMATRIX1                               \
  7368, 10000,  -2141, 10000,  -598 , 10000, \
 -5621, 10000,  13254, 10000,  2625 , 10000, \
 -1418, 10000,   1696, 10000,   5743, 10000

also try with this matrix:
#define cam_CalibrationIlluminant1 21 // D65

For current SX10 matrix, reds are oversaturated (like to many other cameras: SX100, A720, S5IS).
With this matrix, reds are slightly undersaturated.
Violet colors in both matrix are more violet than in jpeg, but they (in dng) are more similar to the real colors (due to  suppression of chromatic aberrations in camera jpeg ?).

[snc]

Did you give it a shot with NIP2? There is a nice tutorial here: http://www.vips.ecs.soton.ac.uk/index.php?title=Colour_calibration_with_nip2&printable=yes

Attached see the image I quickly did with it, and your sample pic. It shows the color matrix it calculated, I suspect it could be used for our purpose as well. Just I didn't have time to look into it yet. Either the values need to be ordered correctly, or the matrix has to be inversed for the transformation... Suspect that a value like 107.07 would be 1.0707 in "our" terms... that would make sense, judging from the scale of them.

[whoever]

Did you give it a shot with NIP2? There is a nice tutorial here: http://www.vips.ecs.soton.ac.uk/index.php?title=Colour_calibration_with_nip2&printable=yes
...

For this to work, you need the real G-M color-checker, not an on-screen fake (plus you need a standard light source). Overall, to calibrate anything, you need a calibrated source. (ArtDen's logic in dng4ps2 is something different from calibration, even though it may produce useful results.)

[snc]

well, you are right with that, of course. but as i said, it was merely a quick test. what i am trying to work out with this is some principle that i could apply. the test image that was processed there surely should give us some "close enough" result as proof-of-concept, should it not?
and really, i am willing to get a color checker chart - it's not like they are hellishly expensive, after all. plus, i have the access to standard light sources and additional color charts.