Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: Colour space in DPP  (Read 12199 times)
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #20 on: July 26, 2009, 05:22:37 PM »
ReplyReply

Quote from: Czornyj
But the real, native color space of the camera may change every time you make a picture
The color space of the camera is given by the spectral responses of the sensels, thus it is fixed.

Quote
camera profiles are describing the camera behavior only in synthetic circumstances, so we can't really say that camera profiles are RAW profiles
What "camera profiles" do you mean? There is no generic camera profile; the profiles are raw processing specific. The same camera can be described by different profiles.

For example what is synthetic in ACR's camera profile relating to a specific camera?

Quote
We have different camera profiles for different occasions - for example modern Nikon Cameras have 9 different camera profiles, and each of these profiles has 20 variants for different illuminants - that makes 180 camera profiles! And in my opinion we can't say that any of these profiles is a "real RAW profile"
1. There is no point to distinguish between "camera profile" and "raw image profile". The raw image data is the incarnation of the camera's image characteristics.

2.These are neither "raw profiles", nor "camera profiles"; they are a combination of camera profiles and custom transformations. I don't know how to name them. The same with Adobe's profiles, particularly the DNG profiles.

Anyway, think of this: one can create such colors with these profiles, which are not reproducable by the camera (like "false colors"). Do you think, that these profiles really expand the gamut of the camera?

Quote from: Jeremy Payne
Certainly changes with different lens and filter combinations.
Do you agree, that the effect of lens and filter is equivalent to changing the illumination? (The same filter could be on the lamps!)

In effect you are saying, that the camera's characteristics depend on the illumination. This is equivalent to stating, that the camera's characteristics depend on the depicted objects. Do you think that the camera's color space is narrower when capturing a checker board than when capturing a colorful scenery?
Logged

Gabor
Jeremy Payne
Guest
« Reply #21 on: July 26, 2009, 05:28:30 PM »
ReplyReply

Interesting ... I guess it depends on whether the "camera" is the sensor ... or the sensor + the rest of the junk like the lens ... no?

But I see your point.

I guess it is easiest to think of the spectral response characteristics of the sensor as the base upon which all else is built and call that the native "space" of the camera.
Logged
Czornyj
Sr. Member
****
Offline Offline

Posts: 1445



WWW
« Reply #22 on: July 26, 2009, 05:43:13 PM »
ReplyReply

Quote from: Panopeeper
The color space of the camera is given by the spectral responses of the sensels, thus it is fixed.
The spectral responses are lens, filter, and illuminant dependend. The color space of camera will be different every time you'll change one of these factors.


Quote from: Panopeeper
What "camera profiles" do you mean? There is no generic camera profile; the profiles are raw processing specific. The same camera can be described by different profiles.

For example what is synthetic in ACR's camera profile relating to a specific camera?
The illuminant it's counted for.

Quote from: Panopeeper
1. There is no point to distinguish between "camera profile" and "raw image profile". The raw image data is the incarnation of the camera's image characteristics.

2.These are neither "raw profiles", nor "camera profiles"; they are a combination of camera profiles and custom transformations. I don't know how to name them. The same with Adobe's profiles, particularly the DNG profiles.

Anyway, think of this: one can create such colors with these profiles, which are not reproducable by the camera (like "false colors"). Do you think, that these profiles really expand the gamut of the camera?
I think, that all colors created with these profiles are "false colors" - the only way to make real "RAW profiles" is to make on-the-fly profile for each RAW image. It would be possible to make such profiles if we had spectral sensitivity of camera+lens+filter and the spectral properties of the scene.
« Last Edit: July 26, 2009, 05:47:51 PM by Czornyj » Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2882



« Reply #23 on: July 26, 2009, 08:13:38 PM »
ReplyReply

Quote from: Czornyj
But the real, native color space of the camera may change every time you make a picture. And camera profiles are describing the camera behavior only in synthetic circumstances, so we can't really say that camera profiles are RAW profiles. We have different camera profiles for different occasions - for example modern Nikon Cameras have 9 different camera profiles, and each of these profiles has 20 variants for different illuminants - that makes 180 camera profiles! And in my opinion we can't say that any of these profiles is a "real RAW profile".


Carrying your argument to its logical conclusion via the process of reductio ad absurdum, then we could not render any raw file since we would likely not have the correct profile. An ICC camera profile made by various profiling software using a multicolored target such as the Color Checker SG usually employs lookup tables and is useful only under carefully controlled conditions such as might be found in a reproduction studio. For general use in field photography such profiles are not all that useful and a more general method is needed.

The approach taken by many raw converters such as DCRaw and Adobe Camera Raw is to use a 3 x 3 matrix conversion to convert from your non-existent camera space to CIE XYZ. Chapter 6 of the Adobe DNG specification, Mapping Camera Color Space to CIE XYZ Space, describes how this done. Since the color filters in a Bayer array sensor are not linear across the range of the human retinal cones, the result will only be a first order approximation. See this quote from Thomas Knoll. Denying the existence of a "camera" space negates this approach. Also, trying to differentiate between a raw file space and the camera space is only obfuscation. As Gabor explained, the space in a raw file is that of the camera from which it was derived.

The newest DNG profiles also provide hue/saturation mapping tables. Accurate color may be very difficult to obtain with this approach, but pleasing colors are usually achieved.

Bill



Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #24 on: July 26, 2009, 10:16:21 PM »
ReplyReply

Quote from: Czornyj
The spectral responses are lens, filter, and illuminant dependend. The color space of camera will be different every time you'll change one of these factors
One could adopt this approach, but it would not be very useful, for a raw processor would have to create lots of profiles for every camera. How many combinations of temperature and tint do you think would be required? For example ACR supports temperature from 2000 K to 10000 K in 50K steps, that is 161 different values, and the tint can be any integer from -150 to +150. That is 161*301, 48461 different combinations.

It is much more useful to describe the sensor's response on its own and account for the illuminant (including lens and filter) through white balancing, an independent step.

Quote
The illuminant it's counted for
No, it is not. The ACR camera profile describes the color transformation between the camera's color space and CIE XYZ relating two illuminants; the actual transformation will be interpolated based on the actual illuminant relative to the two specified illuminants. The two illuminants are typically "Standard light A" (incandescent) and D65 (6504 K natural daylight), but a program converting a native raw file in DNG can choose different basis illuminants (this occurs with the raw files of certain MFDBs).

It is permitted to specify only a single illuminant, but that is useful only if the transformation is specified based on that particular illuminant; this is the normal case with scanners but not with digital cameras, though some MFDB raw->DNG converters require the specification of the actual illuminant and write the transformation fixed for that illuminant (in effect they anticipate the white balancing).
Logged

Gabor
Czornyj
Sr. Member
****
Offline Offline

Posts: 1445



WWW
« Reply #25 on: July 27, 2009, 03:53:05 AM »
ReplyReply

Quote from: bjanes
Since the color filters in a Bayer array sensor are not linear across the range of the human retinal cones, the result will only be a first order approximation.

That's exactly what I meant. I don't deny the existance of camera color spaces, I only think they are more or less virtual, and wouldn't name them "RAW color spaces". Each RAW file has a "real" color space, that is a unique result of camera's spectral sensitivity and scene's illuminant spectra combination. As long as we don't know the spectral properties of scene's light, we can't talk about a "real" RAW color space.
« Last Edit: July 27, 2009, 04:00:02 AM by Czornyj » Logged

Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad