Ad
Ad
Ad
Pages: « 1 2 3 [4]   Bottom of Page
Print
Author Topic: Accurate Colors  (Read 33414 times)
papa v2.0
Full Member
***
Offline Offline

Posts: 194


« Reply #60 on: July 15, 2009, 09:26:39 AM »
ReplyReply

Hi Guys

Ive been following this thread with interest. It looks like there is some grey ares in defining 'accurate' and pleasing etc.

Ive been working in colour appearance (based on our perception)  as opposed to "accurate colormetric colour" (if that the right term!)

Quote
Yes and yes but it could be expensive. Eric Walowit who's on the ICC Digital Camera group has for years proposed the idea that our cameras would record the spectral properties of the scene, embed this as EXIF data with the Raw and, along with the important spectral sensitivity of the camera, build an on-the-fly profile for each image. It makes a lot of sense when technology affordable and dismisses Jack's idea about "one profile for all scenes" but that's admittedly based on the current and not very robust solutions offered today.


I finished some interesting work last year based on the above ideas by Eric Walowit who gave us a presentation in London in 2007t. I introduced a colour appearance model (CIECAM02) into the raw digital pipeline.  The camera was spectrally characterised using a double monochromator and a matrix produced using Erics methods.

The colour appearance model uses CIE colorimerty to deternmine parameters etc but then mathematicly models human perception.
Camera Nikon D70s was used as Eric had spectral data for that camera type.
To cut a long story short the preocess involved obtaining RAW file using DCraw, applying an optimised camera matrix (using Erics methods) to obtain scene referred colorimetry and the applying CIECAM02 and then rendering to sRGB (easiest option for evaluation purposes).

I obtained excellent results and compared with camera JPEG there was a real difference. The scene was more as I remembered it.  I tried to replicate the results using ACR and NIkon Raw developer. Got nearly there but could mot match CIECAM02 for improvement in colour.

Ill post some examples if anyone is interested.

Colour appearance processing is the way forward but there are a few issues still needing ironed out first, but first results are promising.

I hope to publish a paper later this year at Digital Futures 2009 in London
digital futures

Iain






Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 8581



WWW
« Reply #61 on: July 15, 2009, 09:31:47 AM »
ReplyReply

Quote from: papa v2.0
Hi Guys
Ill post some examples if anyone is interested.

Colour appearance processing is the way forward but there are a few issues still needing ironed out first, but first results are promising.

I hope to publish a paper later this year at Digital Futures 2009 in London
digital futures

Iain

Please do and let us know when the paper is available.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
Czornyj
Sr. Member
****
Offline Offline

Posts: 1369



WWW
« Reply #62 on: July 15, 2009, 10:08:23 AM »
ReplyReply

Quote from: papa v2.0
The colour appearance model uses CIE colorimerty to deternmine parameters etc but then mathematicly models human perception.
Camera Nikon D70s was used as Eric had spectral data for that camera type.
To cut a long story short the preocess involved obtaining RAW file using DCraw, applying an optimised camera matrix (using Erics methods) to obtain scene referred colorimetry and the applying CIECAM02 and then rendering to sRGB (easiest option for evaluation purposes).

I obtained excellent results and compared with camera JPEG there was a real difference. The scene was more as I remembered it.  I tried to replicate the results using ACR and NIkon Raw developer. Got nearly there but could mot match CIECAM02 for improvement in colour

Fascinating!
Could you foresee, if the introduction of CIECAM02 in ACR/LR or Nikon Capture alone would help to achive comparable results, or would it also be absolutely necessary to optimise color matrices using Erics methods?
Logged

papa v2.0
Full Member
***
Offline Offline

Posts: 194


« Reply #63 on: July 15, 2009, 11:20:05 AM »
ReplyReply

Quote from: Czornyj
Fascinating!
Could you foresee, if the introduction of CIECAM02 in ACR/LR or Nikon Capture alone would help to achive comparable results, or would it also be absolutely necessary to optimise color matrices using Erics methods?


I think that to use CIECAM02 effectively requires accurate spectral characterization of the camera (which should include lens and any filter.  I performed characterization using the lens supplied with the Nikon with a skylight filter in place. I dare say changing components will have some effect but how much I dont know).  

Methods of calculating a matrix are published in ISO 17321-1:2006 . I dare say that other ways of optimising the matrix can be evolved and incorporated into a RAW workflow. The target based characterization used by ACR at the moment might not be sufficient to produce the required results. I also did a target based characterisation and obtained a matrix but found the approach not as flexible as the spectral based method. So there may be issues there.

Camera manufactures who have accurate spectral characterisations of their products would be in a very good position to develop appearance processing. As for third party developers, this would require a bit more work, but not impossible!

But having said that I cant see it being too long before the idea of CIECAM02 processing catches on. (its in Windows Vista) There is still a bit of work to do as i said to get a robust camera to print workflow. Main stumbling blocks is defining the viewing conditions of the scene and those of the output. CIECAM02 requires these to be defined accurately for optimal results. I used the output conditions defined for a sRGB display (ISO 3664:2009).

I think that there are ways of incorporating the necessary information into an ICC profile, but then arises the need for smart CMMs (another issue altogether).

Using CIECAM02 for general outdoor photography relies heavily on the white point estimation. Using CIECAM02 in a controlled studio environment where the white point can be measured accurately would be ideal. And if the output viewing conditions were controlled  also (in the tube system for example, paper media and emissive displays under uniform lighting spec) a well controlled appearance chain could be set up.
Studio photography would benefit greatly from the use of appearance models as would consumer level cameras where sRGB JPEG is the output referred medium.

Logged
Czornyj
Sr. Member
****
Offline Offline

Posts: 1369



WWW
« Reply #64 on: July 15, 2009, 11:53:53 AM »
ReplyReply

Quote from: papa v2.0
I think that to use CIECAM02 effectively requires accurate spectral characterization of the camera (which should include lens and any filter.  I performed characterization using the lens supplied with the Nikon with a skylight filter in place. I dare say changing components will have some effect but how much I dont know).  

Methods of calculating a matrix are published in ISO 17321-1:2006 . I dare say that other ways of optimising the matrix can be evolved and incorporated into a RAW workflow. The target based characterization used by ACR at the moment might not be sufficient to produce the required results. I also did a target based characterisation and obtained a matrix but found the approach not as flexible as the spectral based method. So there may be issues there.

Camera manufactures who have accurate spectral characterisations of their products would be in a very good position to develop appearance processing. As for third party developers, this would require a bit more work, but not impossible!

But having said that I cant see it being too long before the idea of CIECAM02 processing catches on. (its in Windows Vista) There is still a bit of work to do as i said to get a robust camera to print workflow. Main stumbling blocks is defining the viewing conditions of the scene and those of the output. CIECAM02 requires these to be defined accurately for optimal results. I used the output conditions defined for a sRGB display (ISO 3664:2009).

I think that there are ways of incorporating the necessary information into an ICC profile, but then arises the need for smart CMMs (another issue altogether).

Using CIECAM02 for general outdoor photography relies heavily on the white point estimation. Using CIECAM02 in a controlled studio environment where the white point can be measured accurately would be ideal. And if the output viewing conditions were controlled  also (in the tube system for example, paper media and emissive displays under uniform lighting spec) a well controlled appearance chain could be set up.
Studio photography would benefit greatly from the use of appearance models as would consumer level cameras where sRGB JPEG is the output referred medium.

Thanks a lot for interesting answer, I hope I'll see CIECAM02 processing in Adobe apps soon.

As for spectral characterisation - I wonder if it would be feasible, that some manufactrer would offer inexpensive monochromators for personal use - or maybe it would be too complicated/expansive for mass production?
« Last Edit: July 15, 2009, 11:55:30 AM by Czornyj » Logged

digitaldog
Sr. Member
****
Offline Offline

Posts: 8581



WWW
« Reply #65 on: July 15, 2009, 01:34:56 PM »
ReplyReply

Quote from: Czornyj
Thanks a lot for interesting answer, I hope I'll see CIECAM02 processing in Adobe apps soon.

As for spectral characterisation - I wonder if it would be feasible, that some manufactrer would offer inexpensive monochromators for personal use - or maybe it would be too complicated/expansive for mass production?

My limited understanding from Eric and others is, the spectral data is key here, the CIECAM02 less so (but very, very useful as its based on appearance modeling).
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
papa v2.0
Full Member
***
Offline Offline

Posts: 194


« Reply #66 on: July 15, 2009, 02:08:46 PM »
ReplyReply

Here is one I did earlier

Images have been taken on the NIKON D70s set to automatic mode – ‘point and shoot’
– with no exposure compensation used in any of the scenes.

The file has three images contained in it.
Ist layer is Nikon JPEG as shot.
2nd is my attempt to recreate scene using ACR from .NEF file
3rd is CIECAM02 processed from .NEF file, with this layer is a levels adjustment. The processed files were coming out of CIECAM02 dark. I did not introduce  exposure rendering into the algorithm before converting to sRGB, opting to adjust in photoshop. (dont ask, had a few issues in this area but more interested in what CIECAM02 was doing colour wise). There is no other post processing involved.

So adjust levels to suit - I have my levels marked at the side - it would be interesting if I can get what your optimum level adjustments are.

I viewed this in dim environment on a sRGB monitor with a white point of D65, with the imaged toggeled to a black background in photoshop and all but the levels palette showing at the side.

Please give yourself time to adapt to the dim environment then toggle between layers to see the effect.

Woods near Inverness Scotland.

CAMERA SETTINGS  • ISO 1600  •  f 22  •  1/30s  •  18mm WB 4650
CIECAM02  •  La 500  •  Yb 20  •  ‘avg’  •  Matrix D55 La 780

This is a typical Highland woodland scene. It was taken on a very bright day causing the
leaves to be backlit. There is not much white in the scene and the luminance levels in the
actual wood have fallen to about 500cd/m2.  CIECAM02 has again has opened up the
shadows and has definitely increased overall colourfulness. This is especially noticeable in the
small areas of sky.

See what you think. Ill dig out some more.

Couldnt get pic to attach, dont think it likes tiffs so ive put int on my website
sample ciecam02

« Last Edit: July 15, 2009, 07:15:07 PM by papa v2.0 » Logged
madmanchan
Sr. Member
****
Offline Offline

Posts: 2101


« Reply #67 on: July 15, 2009, 06:00:07 PM »
ReplyReply

CIECAM02 is often used internally when building profiles (i.e., as the space in which residual errors are minimized). See ProfileMaker, for example.
Logged

Czornyj
Sr. Member
****
Offline Offline

Posts: 1369



WWW
« Reply #68 on: July 16, 2009, 03:56:51 AM »
ReplyReply

Quote from: papa v2.0
See what you think. Ill dig out some more.

Couldnt get pic to attach, dont think it likes tiffs so ive put int on my website
sample ciecam02

It looks very promising, indeed. The colors are vivid, and hues are very well separated. I like the way the lights were rendered - the moss and leaves in sun stains kept intensive colors, and well defined details. I'd be glad to see some other examples.

Quote from: madmanchan
CIECAM02 is often used internally when building profiles (i.e., as the space in which residual errors are minimized). See ProfileMaker, for example.

I've also noticed mysterious "CIECAM02" here and there, for example in basICColor display, and Argyll CMS.
A while ago I also found an interesting plug-in for Photoshop (freeware, Windows only):
http://cliff.rames.googlepages.com/ciecam02plugin
« Last Edit: July 17, 2009, 03:14:47 AM by Czornyj » Logged

papa v2.0
Full Member
***
Offline Offline

Posts: 194


« Reply #69 on: July 16, 2009, 08:31:41 PM »
ReplyReply

hi Czornyj

it was a first pass thru as to speak.I would say I could perhaps improve, but the exercise was to see what CIECAM02  could do rather than get a first "bang on" result. it takes a bit of getting  used to the nuances of the input parameters and their subtle influences on the final result.
not bad for a first attempt!

I chose this scene as i wanted to see how the model could cope with a difficult lighting situation, especially with the shadows.
Logged
sandymc
Full Member
***
Offline Offline

Posts: 248


« Reply #70 on: July 17, 2009, 02:57:21 AM »
ReplyReply

Iain,

I'm curious about the details of your method. You say that "applying an optimised camera matrix (using Erics methods) to obtain scene referred colorimetry and the applying CIECAM02 and then rendering to sRGB (easiest option for evaluation purposes)". It seems to me that that is in effect no different to, as Eric Chan says above, using CIECAM02 to generate a profile, then using that profile. Or am I missing something?

Sandy
Logged
papa v2.0
Full Member
***
Offline Offline

Posts: 194


« Reply #71 on: July 17, 2009, 05:55:57 AM »
ReplyReply

Quote from: sandymc
Iain,

I'm curious about the details of your method. You say that "applying an optimised camera matrix (using Erics methods) to obtain scene referred colorimetry and the applying CIECAM02 and then rendering to sRGB (easiest option for evaluation purposes)". It seems to me that that is in effect no different to, as Eric Chan says above, using CIECAM02 to generate a profile, then using that profile. Or am I missing something?

Sandy

I think what Eric Chan means is that the gamut mapping / rendering is done in CIECAM02 appearance coordinates of JCH (lightness, Chroma and Hue) as this is a perceptually better space than CIELAB.

What can happen is that XYZ is converted to JCH and back to XYZ with the same input and output CIECAM02 parameters, so in effect no actual appearance processing is taking place, only a conversion to another colour space.


And yes the scene referred colorimetry and the viewing conditions can be saved into a profile for later processing in situations where the output viewing conditions and device are not known. There are two tags in the ICC v4 profile spec (page96) for viewing conditions:
viewingCondDescTag - viewing condition description
viewingConditionsTag - Viewing condition parameters.

So the framework is there, just needs to be put into use.

Ps added another sample pic  rainbow over Crystal Palace London
« Last Edit: July 17, 2009, 07:06:33 AM by papa v2.0 » Logged
Andrew Fee
Jr. Member
**
Offline Offline

Posts: 87


WWW
« Reply #72 on: July 17, 2009, 12:38:12 PM »
ReplyReply

I've not really been following all of this topic as it's been a bit over my head, but looking at your images there seems to be a few major downsides to the CIECAM02 rendering.

There seems to be significantly more chroma noise, an artefact that looks like really bad chromatic aberration around fine details, and looking at the histogram is a bit concerning. I'm also about curious—are the “ACR from NEF” layers using a custom profile, or just one of the standard ones? (I found that none of the Adobe ones were any good with my camera)
Logged
Czornyj
Sr. Member
****
Offline Offline

Posts: 1369



WWW
« Reply #73 on: July 17, 2009, 01:35:53 PM »
ReplyReply

Quote from: Andrew Fee
I've not really been following all of this topic as it's been a bit over my head, but looking at your images there seems to be a few major downsides to the CIECAM02 rendering.

There seems to be significantly more chroma noise, an artefact that looks like really bad chromatic aberration around fine details, and looking at the histogram is a bit concerning. I'm also about curious—are the “ACR from NEF” layers using a custom profile, or just one of the standard ones? (I found that none of the Adobe ones were any good with my camera)

You should take into account, that - like usual in such cases - Iain used the open source DCRAW converter as a base to further experiments, so chroma noise and chromatic aberration have nothing to do with CIECAM02 (it's only a color model), but it's simply a matter of demosaicing algorithm of DCRAW (that doesn't handle noise and aberration as good as commercial Nikon or Adobe products).

Apart from noise etc., the samples look really interesting - the colors are clean, vivid and well separated, even in the shadows and highlights areas. To my eye these pictures look atractive, but also suprisingly realistic.
« Last Edit: July 17, 2009, 01:56:10 PM by Czornyj » Logged

papa v2.0
Full Member
***
Offline Offline

Posts: 194


« Reply #74 on: July 17, 2009, 02:08:37 PM »
ReplyReply

Quote from: Andrew Fee
There seems to be significantly more chroma noise, an artefact that looks like really bad chromatic aberration around fine details, and looking at the histogram is a bit concerning. I'm also about curious—are the “ACR from NEF” layers using a custom profile, or just one of the standard ones? (I found that none of the Adobe ones were any good with my camera)

Hi Andrew. Any noise in the original RAW will be accentuated by CIECAM02. Noise is not a CIECAM02 artifact.  I used DCRAW as it exports demosaiced RAW files, my starting point. I cant get this from NIkon or ACR.

The ACR from NEF layer was my attempt to replicate the scene using ACR. All adjustments were were manual, no profiles.
Logged
Pages: « 1 2 3 [4]   Top of Page
Print
Jump to:  

Ad
Ad
Ad