Dan Margulis Applied Color Theory - Linear Gamma (Gamma 1.0)

Date: Sat, 21 Dec 2002 12:08:44 -0500
From: Terry Britton
Subject: Linear color space from AIM

Everyone,
I have been experimenting with Accurate Image Manipulation's (AIM's) linear color space for gaussian blurs and UnSharpMask applications, ( http://www.aim-dtp.net/aim/evaluation/gamma_error/index.htm ) and it seems to work as advertised, avoiding the color changes in the blurs as well as 'white edges' and color changes in USM. I load in an image and apply Adobe 1998 (now using the Microsoft ICM as per AIM's recommendation for it's better slope control - http://www.aim-dtp.net/aim/photoshop/v6/slope-limiting/index.htm ). I then use 'convert to profile' to change to the AIMRGBPro linear color space, and after I'm done with the filters, I convert back again to Adobe 1998 with no detectable loss. (The image of the Kodak Q-60 including the icc profiles for the linear color space is here: http://www.aim-dtp.net/aim/evaluation/gie/small-q60-adobergb.zip )

I've tried this in several situations, and it really does improve the quality of USM quite dramatically, to my eyes. Of course, avoiding the color changes caused by blending errors elsewhere (in blurs, strokes, and blending modes) is very nice. (I understand that Adobe included the last checkbox in the Advanced version of the Color Settings to compensate for these color errors during antialiasing...especially with stroked paths. But that check box does nothing for blurs, nor for other filters that I know of.)

Has this been discussed in archives (where would I look?) or are there issues or opinions among the readers of this forum which I should be taking into consideration? Any discussion would be welcome and interesting! (Actually, any discussion of this entire website would be welcome and interesting!)

Thank you.

Terry Britton

Date: Sun, 22 Dec 2002 01:28:58 -0000
From: Stephen Marsh
Subject: Re: Linear color space from AIM

> I have been experimenting with Accurate Image Manipulation's (AIM's)
> linear color space for gaussian blurs and UnSharpMask applications,
> ( http://www.aim-dtp.net/aim/evaluation/gamma_error/index.htm )
Hi Terry, this is a 'hot subject' and it has been beaten to death on many lists over many years.

In the end it racks up as pretty much all segments of the graphics industry from Adobe programmers to 'industry names' are against gamma 1 work space edits - while Timo and one other programmer that I know of bother with linear edits. History may prove Timo right and everyone else wrong - it has happened before. At this point though, I do not think I would be putting my money on the gamma 1 approach for general edits, although in some cases there may be a minor advantage for some of the image content.

The other programmer who seems to employ gamma1 can be found in the free Panorama Tools plugs for Mac or PC (excellent plugs, not user friendly though):

http://home.no.net/dmaurer/~dersch/gamma/gamma.html

This is a more balanced approach though - the author notes that it is more for CG than for scans/captures:

>> "Unfortunately, gamma correction and the reverse transformation are not completely lossless. Due to the finite Bitdepth there are always some quantization errors which might add up if you chain many transformations. In practice it is a trade-off between two evils. For computer rendered images, a degamma/gamma correction is almost always adequate, for real-world photographs it hardly matters."

> ...(now using the Microsoft ICM as per AIM's
> recommendation for it's better slope control -
> http://www.aim-dtp.net/aim/photoshop/v6/slope-limiting/index.htm ).

What about the Mac users wanting to use this technique? By MS CMM I guess you mean Heidelgerg/Linocolor, so perhaps Mac users could use that to see what Timo is talking about, if they wanted to.

It was my understanding that differences between most CMM were only very minor and cosmetic, I may have to look into this deeper.

> I've tried this in several situations, and it really does improve the
> quality of USM quite dramatically, to my eyes.

Try a 1.8 or 2.2 gamma space and sharpen the gamma image the same as the linear image (same values in USM) - but on a duped layer. Then dupe the layer again so you have two USM layers. Set one to lighten and the other to darken. Reduce the opacity of the lighten and or darken layers. The results of this can then be blended in luminosity mode into the original data, so you have both light/dark halo intensity and luminosity blending.

Extreme gamma spaces (1 or 3) bias different extreme tonal ranges.

> Has this been discussed in archives (where would I look?)

http://groups.yahoo.com/group/colortheory

Then use suitable keywords in the search archives section.

> or are there
> issues or opinions among the readers of this forum which I should be
> taking into consideration? Any discussion would be welcome and
> interesting! (Actually, any discussion of this entire website would be
> welcome and interesting!)

What a loaded question!

Take care, work on dupes - keep an open mind but also keep a big bag of salt handy.

Bruce Lindbloom has some 'perfect' CG images which show the issues with linnear gamma and reg/hi big edits on CG images (which may not hold true to scans/captures - but it does demonstrate a point):

http://www.brucelindbloom.com/index.html?ReferenceImages.html

Perhaps the best place for me to point you to for some balance to Timo's site are these links below. Adobe have axed the archive on their site where Timo and Bruce Fraser 'discussed' (for want of a better term) the subject.

http://www.vision.ee.ethz.ch/~buc/brechbuehler/mirror/color/GammaFAQ.html#RTFToC17

http://www.vision.ee.ethz.ch/~buc/brechbuehler/mirror/color/GammaFAQ.html

http://www.faqs.org/faqs/graphics/colorspace-faq/

The questions I like to ask when all this 'gamma editing' hoohah pops up is - what about LAB and CMYK edits??? LAB has been noted as having a gamma close to 2.5 or perhaps 3, although I am not sure if LAB has a 'gamma' as such, it must be hidden from users like the white point is, LAB is usually considered LAB with no further qualification (although there are different LAB versions using D50 or D65). So if gamma 1 or gamma 2.2 edits are supposed to be so much better - what happens when you edit in LAB or CMYK or even gray with a dot gain profile instead of gamma?

I have not received an answer to this yet. From my tests I know what my conclusions are - but with all the 'religious' hype that surrounds gamma and RGB spaces it gives me a chuckle that other modes are not given the same attention.

And on the 2.2 'perceptually' uniform gamma issue - I wonder why some of the wide gammut spaces out there use 1.8 gamma, if 2.2 is so much better?

Hope this helps,

Stephen Marsh.

Date: Sun, 22 Dec 2002 01:54:56 -0000
From: Stephen Marsh
Subject: Re: Linear color space from AIM

I meant to mention the very handy levels calc. offered by Bruce Lindbloom:
http://www.brucelindbloom.com/LevelsCalculator.html

2.2 source image converted to gamma 1.0: 256 levels reduced to 184 levels in an 8 bpc transform.

2.2 source image converted to LAB: 256 levels reduced to 234 in an 8 bpc transform.

This does/may not reflect any apparent _visual_ change to the file on the monitor or on output - just what is going on under the hood, so to speak. Since gamma 1 is a huge curve to pull, you loose a lot of levels on 8 bpc data moving from the gamma corrected source image to the linear working space, then you have to convert to a gamma compensated or dot gain compensated space at a later point in the files life, as well as all the edits that take place in the working space. Using high bits avoids all this of course, but that's another 'golden eye' debate which has been kicked over many times too, just like the LAB mode change errors.

Also compare the separate channels of a 'regular' gamma file against a linear gamma 1 converted file - then ask yourself which work space file you would prefer to base channel blends or grayscale mixing conversions off. While you are at it, why not compare a gamma 3 space as well, being the opposite extreme of gamma 1, as well as the 2.2 and 1.8 'standards'.

Stephen Marsh.

Date: Sat, 21 Dec 2002 20:25:38 -0700
From: Andrew Rodney
Subject: Re: Re: Linear color space from AIM

Stephen Marsh wrote:
> History may prove Timo right and
> everyone else wrong - it has happened before

It hasn1t happened in almost 13 years since Photoshop first shipped. I think we can all go on with our digital lives and find something far more interesting to debate or worry about. Timo is somewhat of a joke to the 10th floor at Adobe (where all the Photoshop engineers hang). Adobe (and a lot of other digital imaging companies) have been doing just fine using their process and as yet, I don1t know a single company that has Timo working for them. Of course, perhaps the earth really is flat...

If you ever run into Chris Cox, mention Timo and watch the expression on his face. It's hilarious.

> It was my understanding that differences between most CMM were only
> very minor and cosmetic, I may have to look into this deeper.

Unless you hit a bug, that's my understanding too. And when you do hit one, you'll know it in a second because your output will look like poop. I think Bruce Fraser did a test of the various CMM's when they all behaved and found the differences was at worse case 1% in a single color channel. Nothing to lose sleep over.

Andrew Rodney

Date: Sat, 21 Dec 2002 23:09:19 -0700
From: Chris Murphy
Subject: Re: Re: Linear color space from AIM

Yeah I agree with Stephen that gamma 1.0 edits, especially in 8-bits, is probably not a good idea especially if there is appreciable and important shadow detail. However, there are some "advantages" colorimetrically speaking, of blends being based on gamma 1.0. There is an option in Color Settings with the advanced mode checked for exactly that. So while the image is still in a gamma 1.8 or 2.2 space, RGB blending occurs with gamma 1.0.

An easy example is this: what color do you get when you add red light and green light? You should get yellow. Take the paint brush tool and draw some green and then draw some red through it and look at the transition area. It's dirty. Now go change the rgb blending gamma in Color Settings and, say OK, then in the same document draw some more red through another green area and you'll see yellow in the transition area.

Chris Murphy
Color Remedies (TM)
www.colorremedies.com

Date: Sun, 22 Dec 2002 12:13:39 -0700
From: Chris Murphy
Subject: Re: Re: Linear color space from AIM

Andrew Rodney wrote:
> Unless you hit a bug, that's my understanding too. And when you do hit one,
> you'll know it in a second because your output will look like poop. I think
> Bruce Fraser did a test of the various CMM's when they all behaved and found
> the differences was at worse case 1% in a single color channel. Nothing to
> lose sleep over.

That test was based on Photoshop 5 or 6, both of which had settings that were overridden by the ColorSync control panels Preferred CMM setting. Last time I talked to Bruce, during PS 7 beta testing, about this, he said he thought he'd need to redo the test. My own experience is that the CMMs can cause a good bit of difference. Usually it is only 1-2% but sometimes I've seen a lot more than that - even addition of hue casts. For example if you convert something with the Kodak CMM and Agfa CMM, compared to each other the Kodak CMM conversion is cooler and the Agfa one is warmer.

Chris Murphy
Color Remedies (TM)
www.colorremedies.com

Date: Tue, 24 Dec 2002 13:57:14 -0500
From: Terry Britton
Subject: Re: Gaussian blur and gamma

I think the strongest argument for limited, occasional use of Timo's linear color space would be in the instance of using Gaussian blur, which is one blending area not addressed by Adobe's "Advanced Mode Gamma 1" checkbox (Though I'm not sure why the exclusion happened, it did.)
The example Chris Murphy gives using the Green and Red blending situation (from antialiasing) is a good example of why this Gamma 1 setting is especially useful in graphic arts, but I use Gaussian blur on occasion on a layer as a way of removing paper textures from old photographic prints (I then paint on a light to heavy mask to reveal sharpened features in the layer below) - and for other things. It rather shocked me that certain colors would blend entirely 'wrong' even with the "Advanced Mode Gamma 1" box checked. (I tried the very telling download examples at AIM's Photoshop 6 page here http:/.aim-dtp.net/toshop/nding/blending.htm at http://www.aim-dtp.net/aim/photoshop/v6/blending/rgb_blending.htm )

The author describes this page saying, "This page demonstrates the effect of the Blend RGB Colors Using Gamma 1.0 option and compares the result with a GaussianBlur filter that also blends RGB colors but is not covered by the option, that is the case at least for all the filter operations but possibly for many other editing operations too."

There's no arguing with what happens to the blurred checkerboard, so I began to worry... (I don't want blurs that should produce plum colors turning things gray!)

Terry Britton

Date: Thu, 26 Dec 2002 09:22:48 EST
From: Dan Margulis
Subject: Re: Gaussian blur and gamma

Terry Britton writes,
>I think the strongest argument for limited, occasional use of Timo's linear
>color space would be in the instance of using Gaussian blur...

That's correct, but it has nothing to do with Timo or his theories. Blurs ordinarily give slightly happier results at much lower gammas than we customarily use. It doesn't have to be a 1.0 gamma as such.

The other use for low gamma, as I see it, is in grossly dark images where we have applied a false profile as a corrective measure. Previously, I've said that if we've applied something weird like Wide Gamut RGB at a .8 gamma to try to drastically lighten an image, we should probably exit immediately by converting to LAB or some more conventional RGB. Having redone a lot of my example images, I now think that they benefit from having as much curving, blending, and blurring done in the weird setting before making the conversion as possible. This seems to minimize certain problems.

>...which is one
>blending area not addressed by Adobe's "Advanced Mode Gamma 1" checkbox
>(Though I'm not sure why the exclusion happened, it did.)

It shouldn't be there except perhaps as an option. That "blending" refers to color painting operations only, not channel-by-channel filters or channel blending. It sounds like you're trying to avoid a color shift. You may wish to fade the blur to Luminosity after making it.

While blurring might be better at an ultra-low gamma, sharpening and addition of noise are worse, so having filter operations take place at a low gamma generally is a bad idea.

Dan Margulis

P.S. I do want to warn that Stephen Marsh's earlier remark about this topic "having been beaten to death on other lists" qualifies as the understatement of the year. A combination of religious war and circus might be a better description. We have room for a civil discussion of the impact of extreme gammas here, but not for the type of nonsense that many of us have seen elsewhere.

Date: Thu, 26 Dec 2002 10:10:45 -0800 (PST)
From: Andy Adams
Subject: Gamma Graphing

Since the subject of gamma came up, I thought I would ask something concerning gamma and the graphing of it's curve.
When doing some research on line about ICC profiles and their conversions in and out of the Profile Connection Space (PCS), I came across some information that discussed how Tone Reproduction Curves (TRC) are the curves that define the gamma of the device or color space. Then the site proceeded to show a graph of such a TRC.

My question is basically, how can one graph the gamma. And is a gamma of 1.0 the same as a straight line curve? I could be getting the concept of gamma confused with TRCs, if so let me know. I've been trying to figure this out in my spare time (and there isn't much of it).

Andy Adams

Date: Thu, 26 Dec 2002 22:33:23 -0000
From: Stephen Marsh
Subject: Re: Gamma 1 Gaussian Blurs vs Other Modes

Terry writes:
>> I think the strongest argument for limited, occasional use of Timo's linear color space would be in the instance of using Gaussian blur, which is one blending area not addressed by Adobe's "Advanced Mode Gamma 1" checkbox (Though I'm not sure why the exclusion happened, it did.)<<

Hi again Terry, thank you for giving me yet another Photoshop issue to explore.<g>I only infrequently visit the AIM site and these v6 updates are new to me.

I have noted some things that may have helped for converting lineart or '1 bit' g/scale to either gamma 1 or 3, but even this was not such a big deal that I made it a golden rule or part of my lineart scanning/editing workflow. In the case of a white egg on a white background, polar bear in a snowstorm or other hi key images - then having a gamma between 1-1.4 may be of some help (just as really dark low key image may be better off being edited in a higher gamma space than even 2.2).

The RGB blending means exactly that - where blends/overprints occur then gamma is taken into consideration. Pulling a curve or filtering does not account as a 'blend' so even if using the advanced colour setting for gamma 1 blends many other operations are happening in the RGB work space gamma. So this is why the Gaussian blur example at the AIM site does not work with the advanced colour setting option. It's a fine and confusing point.

>> The example Chris Murphy gives using the Green and Red blending situation (from antialiasing) is a good example of why this Gamma 1 setting is especially useful in graphic arts,<<

From a graphic arts viewpoint - a form of trapping is applied with the non linear gamma compensated work space edit. From an artistic point of view, then I can see why this may not always be considered a good thing and that linear blending is considered 'better' as true edge blends occur.

Then there are the other considerations - the gamma 1 rgb blend in advanced colour settings affects how solid/tints of colours overlay/overprint/blend etc. This may not always be as easy to figure out what is 'correct' - what is right for one image may not be for another, or different parts of one image may be better off with two different gamma blending approaches.

>> It rather shocked me that certain colors would blend entirely 'wrong' even with the "Advanced Mode Gamma 1" box checked.<<

>> There's no arguing with what happens to the blurred checkerboard, so I began to worry...
> (I don't want blurs that should produce plum colors turning things gray!)<<

Download the AIM coloured checkerboard CG...

Yes, the 2.2 gamma (key point, see below) test results in gray when colour is expected. Yes the gamma 1 space results in a saturated colour (perhaps not good but better than gray). Timo does not demonstrate or mention what happens when you do the same or similar tests to 1.8 gamma RGB, LAB and CMYK.

When LAB and 1.8 gamma tests are performed the result is also colour and not gray (a good thing) and for an in-gamut test in CMYK results are different but similar again - colour and not gray. So it would seem at a casual glance that gamma 2.2 is far from 'colorimetrically correct' in this case for this edit - but I do not like the gamma 1 results either...LAB seems to show a pleasing result and although different - the 1.8 and CMYK results are closer to the LAB result than 1.00 which is more saturated.

Regards,

Stephen Marsh.

Date: Fri, 27 Dec 2002 14:40:43 -0500
From: Terry Britton
Subject: Re: Gaussian blur and gamma

Dan Margulis wrote:
>The other use for low gamma, as I see it, is in grossly dark images where we
>have applied a false profile as a corrective measure. Previously, I've said
>that if we've applied something weird like Wide Gamut RGB at a .8 gamma to
>try to drastically lighten an image, we should probably exit immediately by
>converting to LAB or some more conventional RGB. Having redone a lot of my
>example images, I now think that they benefit from having as much curving,
>blending, and blurring done in the weird setting before making the conversion
>as possible. This seems to minimize certain problems.

Thank you for yet another very useful tip. (I've been amazing my friends with the "little known" tilde shortcut trick in your recent article, which in my case has about a trillion applications! Gadzooks that works well!)

>It shouldn't be there except perhaps as an option. That "blending" refers to
>color painting operations only, not channel-by-channel filters or channel
>blending. It sounds like you're trying to avoid a color shift. You may wish
>to fade the blur to Luminosity after making it.

I'm not entirely certain what is meant by that last statement. Could you please clarify that? Is this an extension of your Luminosity channel masking tip?

>While blurring might be better at an ultra-low gamma, sharpening and addition
>of noise are worse, so having filter operations take place at a low gamma
>generally is a bad idea.

Which begs the question - can you work in different gamma spaces for different operations (low for blurring, higher for noise and sharpening?)

>P.S. I do want to warn that Stephen Marsh's earlier remark about this topic
>"having been beaten to death on other lists" qualifies as the understatement
>of the year. A combination of religious war and circus might be a better
>description. We have room for a civil discussion of the impact of extreme
>gammas here, but not for the type of nonsense that many of us have seen
>elsewhere.

Extremely well put!<g>

Terry Britton

Date: Fri, 27 Dec 2002 15:07:39 -0500
From: Terry Britton
Subject: Re: Gaussian blur and gamma

Stephen Marsh wrote:
>I have noted some things that may have helped for converting lineart
>or '1 bit' g/scale to either gamma 1 or 3, but even this was not such
>a big deal that I made it a golden rule or part of my lineart
>scanning/editing workflow. In the case of a white egg on a white
>background, polar bear in a snowstorm or other hi key images - then
>having a gamma between 1-1.4 may be of some help (just as really dark
>low key image may be better off being edited in a higher gamma space
>than even 2.2).

So it doesn't seem like you have contradicted Dan's statement just made, I'm making particular note of Dan's qualification in his recent post, where he said:

"The other use for low gamma, as I see it, is in grossly dark images where we have applied a false profile as a corrective measure. Previously, I've said that if we've applied something weird like Wide Gamut RGB at a .8 gamma to try to drastically lighten an image, ...."

So, however, in normal editing where you are NOT applying a false profile as a corrective measure, it is good to use higher gamma values, correct?

>The RGB blending means exactly that - where blends/overprints occur
>then gamma is taken into consideration. Pulling a curve or filtering
>does not account as a 'blend' so even if using the advanced colour
>setting for gamma 1 blends many other operations are happening in the
>RGB work space gamma. So this is why the Gaussian blur example at the
>AIM site does not work with the advanced colour setting option. It's
>a fine and confusing point.

Aha! I get it. Thank you!

>>From a graphic arts viewpoint - a form of trapping is applied with
>the non linear gamma compensated work space edit.

Now THAT I will have to investigate further. A built in three or so pixel trap? As in, the colors actually separate with that little bit of overlap? So, in graphic arts applications, we should be careful to turn anti-aliasing off if we or our printer are intending to use a dedicated professional trapping program later on down the road?

>From an artistic
>point of view, then I can see why this may not always be considered a
>good thing and that linear blending is considered 'better' as true
>edge blends occur.

I wonder if this blending of green and red to create yellow in the gamma 1 RGB space converts and separates to a yellow "interface" between the two when converting to CMYK and producing separations. Blended green and red inks simply produce muck. Perhaps that is what the default gamma settings are trying to convey?

>Then there are the other considerations - the gamma 1 rgb blend in
>advanced colour settings affects how solid/tints of colours
>overlay/overprint/blend etc. This may not always be as easy to figure
>out what is 'correct' - what is right for one image may not be for
>another, or different parts of one image may be better off with two
>different gamma blending approaches.

Which again begs the question, can there be multiple gamma blending approaches in a single document? And if so, how would that be accomplished?

>When LAB and 1.8 gamma tests are performed the result is also colour
>and not gray (a good thing) and for an in-gamut test in CMYK results
>are different but similar again - colour and not gray. So it would
>seem at a casual glance that gamma 2.2 is far from 'colorimetrically
>correct' in this case for this edit - but I do not like the gamma 1
>results either...LAB seems to show a pleasing result and although
>different - the 1.8 and CMYK results are closer to the LAB result
>than 1.00 which is more saturated.

These are very useful observations. Thank you, Stephen. I will experiment with different gamma settings for different blurring situations a bit and try to arrive at some standard practices for oft-encountered situations to follow when doing these restorations.

Terry Britton

Date: Sat, 28 Dec 2002 11:35:22 -0000
From: Stephen Marsh
Subject: Re: Gaussian blur and gamma

> So, however, in normal editing where you are NOT applying a false
> profile as a corrective measure, it is good to use higher gamma values,
> correct?

Depends on who you ask!<g>

Most will, as you say - edit in gamma compensated spaces of 1.8 or 2.2. Since v5.x ColorMatch RGB has been a mainstay for me, although I have played with A98 but not in a normal RGB work flow.

I was not talking of assigning a profile, but converting from a gamma compensated source space into gamma 1. Two different things. Editing withing that assigned space is different than editing within a file converted to that space too.

> >>From a graphic arts viewpoint - a form of trapping is applied with
> >the non linear gamma compensated work space edit.
>
> Now THAT I will have to investigate further.

This is the dark coloured edge type effect which happens in say gamma 1.8 or 2.2, as demonstrated here:

http://home.no.net/dmaurer/~dersch/gamma/gamma.html

I was not saying that this may be a wanted effect, just that the edge blend in the gamma compensated RGB with no linnear blending will convert to CMYK with a less clean line between the two elements.

It all depends on the user and situation whether this is considered a good thing or not.

> So, in graphic arts applications, we should be careful to turn
> anti-aliasing off if we or our printer are intending to use a dedicated
> professional trapping program later on down the road?

What a minefield!

It is pretty rare for raster elements to be trapped - most software is set-up to ignore CT elements and to only perform traps between vector data. Of course the trapping is probably going on at the raster level and not object - the original raster areas are ignored when the entire file is rasterized for trapping. The old Scitex LW/CT approach is an example of what I mean here.

It all depends on the ink build of the two abutting objects, as well as the press on what type of trap is made (and other variables).

Generally speaking, it is best not to apply trapping anywhere - and leave this to those that know best. If you do know what you are about, but others are going to do this anyway - I would proabably ignore any vector traps (as they will do this) and perhaps only apply them in the _very_ rare raster 'graphic' (as opposed to image). Most images do not need traps, logos, page layout elements and 'graphics' often do (but not always).

> I wonder if this blending of green and red to create yellow in the gamma
> 1 RGB space converts and separates to a yellow "interface" between the
> two when converting to CMYK and producing separations. Blended green and
> red inks simply produce muck. Perhaps that is what the default gamma
> settings are trying to convey?

No, RGB is light based - it is attempting to show an additive process (apart from those gamma compensated edge issues).

> Which again begs the question, can there be multiple gamma blending
> approaches in a single document? And if so, how would that be accomplished?

No.

Separate docs, different gammas or advanced blend setting of 1.00, masks and converting to a common 'assembly' space from the different sources (either work space RGB or LAB).

> These are very useful observations. Thank you, Stephen. I will
> experiment with different gamma settings for different blurring
> situations a bit and try to arrive at some standard practices for
> oft-encountered situations to follow when doing these restorations.

Let the list know your findings.

Regards,

Stephen Marsh.

Date: Sat, 28 Dec 2002 13:56:38 -0000
From: Stephen Marsh
Subject: Re: Gamma Graphing

Andy Adams wrote:
> My question is basically, how can one graph the gamma.
> And is a gamma of 1.0 the same as a straight line
> curve? I could be getting the concept of gamma
> confused with TRCs, if so let me know. I've been
> trying to figure this out in my spare time (and there
> isn't much of it).

Andy, It is my understanding that gamma 1 is displayed as a straight 45 degree line (curve?) - indicating a 'neutral' baseline response.

I just checked in Macromedia xRes which I knew had a gamma curve, which goes from 0 to 5. 0 is indicated as an inverse curve, 1 as linear and higher values bend the curve as I would expect.

I can email a small GIF animation of this display if reqd.

Not sure on the gamma 0 though...I would have thought this is the 'same' as gamma 1 - but Macromedia seem to think differently. I have to confess that the 'laws' behind gamma curves are not something that I am fully conversed in<g>.

Stephen Marsh.

Date: Sat, 28 Dec 2002 15:18:53 EST
From: Dan Margulis
Subject: Re: Gamma Graphing

Andy Adams writes:
>My question is basically, how can one graph the gamma.
>And is a gamma of 1.0 the same as a straight line
>curve? I could be getting the concept of gamma
>confused with TRCs, if so let me know. I've been
>trying to figure this out in my spare time (and there
>isn't much of it).

You wouldn't be able to graph it because you have no meaningful second variable to graph it against. Gamma is nothing more than a fudge factor, a method of defining the darkness of a given RGB file. The idea is to more closely approximate how humans evaluate darkness, but it's a very, very, very primitive way of doing it--we don't see different intensities of red, green, and blue using the same formula, and we don't see uniformly across darknesses anyway.

The best way to treat gamma is as a straight midtone adjustment to an RGB file. This seems a whole lot easier than learning about reciprocal power curves. If, however, you absolutely need to get formulae, try

http://ourworld.compuserve.com/homepages/compuphase/cmetric.htm

or google Charles Poynton Color FAQ.

Dan Margulis

Date: Sat, 28 Dec 2002 15:18:23 EST
From: Dan Margulis
Subject: Re: Re: Gaussian blur and gamma

Terry Britton writes,
>I'm not entirely certain what is meant by that last statement. Could you
>please clarify that? Is this an extension of your Luminosity channel
>masking tip?

You seemed to be indicating that your blurs were subtly changing the colors as well as eliminating detail. If you wish to avoid this effect, you can do the blur on a duplicate layer with the layer mode set to Luminosity. Or, you can, immediately after applying the blur, choose Edit: Fade>100% Luminosity. Either will restore the original colors while allowing the blur.

>Which begs the question - can you work in different gamma spaces for
>different operations (low for blurring, higher for noise and sharpening?)

In effect, you do that by working in LAB. That's part of the reason that LAB moves often seem very effective. I've been doing research on sharpening/blurring over the last months and will be posting findings shortly, as I will be changing certain recommendations.

Dan Margulis

Date: Sat, 28 Dec 2002 13:08:53 -0800
From: Richard Chang
Subject: Gamma Graphing

Andy Adams wrote:

>> My question is basically, how can one graph the gamma.
>> And is a gamma of 1.0 the same as a straight line
>> curve? I could be getting the concept of gamma
>> confused with TRCs, if so let me know. I've been
>> trying to figure this out in my spare time (and there
>> isn't much of it).
Gamma's photographic definition is described as a deviation from 1 (A gamma of 1 is indeed, a linear response). Gamma is a confusing concept because there's no point of reference unless it's linear. It's also obfuscated because our traditional photographic gamma concept has been applied to monitor displays of R,G, and B. Gamma's job is to provide a display condition that mimics the picture data, as it should render on a reflective target, an interesting notion given the differences between transmissive and reflective physics.

That said, gamma is quite valuable as a transitioning tool when making digital captures, it increases shape in the shadow (and smoothens the highlight at the same time). A gamma curve is noteworthy because it builds contrast (shape) in the shadow but gives it away in the highlight; the endpoints don't move.

Richard Chang
www.TransitionOfTone.com

Date: Sun, 29 Dec 2002 16:23:35 -0500
From: David McDowell
Subject: Re: Gamma Graphing

Let's not get carried away with the term gamma. Its two principle definitions are relatively simple and straightforward.
In photography, gamma is usually defined as the slope of the straight line portion of the plot of density vs logE.

i.e., the contrast of a photographic material (within its working range).

Gamma as used in the television industry is the power function that in either a simple or complex definition describes the relationship between gun voltage and intensity (or luminance).

In both of these definitions gamma is a single valued parameter and therefore cannot be plotted unless someone defines a further relationship which can be related back to gamma (i.e., scene density range for a fixed scene luminance range vs film gamma).

David Q. McDowell
Standards Consultant
51 Parkwood Lane
Penfield, NY 14526
Tel: 585-383-1706, Fax: 585-385-3828

Date: Sun, 29 Dec 2002 23:35:51 -0000
From: Gary Owens
Subject: Re: Gamma Graphing

David McDowell wrote:
> In photography, gamma is usually defined as the slope of the straight line
> portion of the plot of density vs logE.

The definition I'm looking at defines gamma as a "measurement in sensitometry to describe the angle of the straight line portion of a characteristic curve of a photographic emulsion. It expresses the degree of development ans is a form of contrast measurement."
> I saw a good example of contrast in the different workspaces the other day when I was reading a review of the Canon D1s. www.dpreview.com.

Kodak's "Quality Enlarging with Kodak B/W Papers" shows some good examples of characteristic curves.

Gary Owens

Date: Sun, 29 Dec 2002 20:24:52 -0500
From: David McDowell
Subject: Re: Re: Gamma Graphing

The primary reference for this and many other photographic terms and concepts is "The Theory of the Photographic Process" edited by T H James. My copy is the 4th Edition dated 1977. It is still the definitive reference for the photographic industry.
David Q. McDowell
Standards Consultant
51 Parkwood Lane
Penfield, NY 14526
Tel: 585-383-1706, Fax: 585-385-3828

Date: Mon, 13 Jan 2003 13:30:17 -0600
From: "Cliff White
Subject: Re: Gamma and color space

This may be a sophmoric question. I use Adobe 98 as my RGB color space before converting to CMYK for magazine output. I use Adobe 98 because it supposedly most closely matches the color space of my EOS D1. I calibrate my monitor at Gamma 1.8. I briefly toyed with 2.2, but it seemed too dark to me. Dan mentioned that Adobe 98 was a 2.2 color space, does this mean that I should calibrate my monitor at 2.2 for more accurate display of images, or am I ok in 1.8? Or, a more broad question, why is a color space gamma specific?
CLIFF WHITE
Staff Photographer
Missouri Department of Conservation
(573)751-4115 ext. 3854
P.O. Box 180
Jefferson City, MO 65102

Date: Mon, 13 Jan 2003 13:31:35 -0700
From: Andrew Rodney
Subject: Re: Re: Gamma and color space

Cliff White wrote:
> This may be a sophmoric question. I use Adobe 98 as my RGB color space
> before converting to CMYK for magazine output. I use Adobe 98 because
> it supposedly most closely matches the color space of my EOS D1. I
> calibrate my monitor at Gamma 1.8. I briefly toyed with 2.2, but it
> seemed too dark to me. Dan mentioned that Adobe 98 was a 2.2 color
> space, does this mean that I should calibrate my monitor at 2.2 for more
> accurate display of images, or am I ok in 1.8? Or, a more broad
> question, why is a color space gamma specific?

The gamma of a working space has nothing to do with the gamma of your display. They only share the word 3Gamma.2 You can calibrate the display to gamma 1.8 and work with a 2.2 gamma editing space just fine inside of Photoshop because it divorces the way we edit files from the condition of the display. A user on a PC could calibrate the display to 2.2 gamma and open the same file you have and see the same preview.

Andrew Rodney

Date: Tue, 14 Jan 2003 13:01:22 -0500
From: Terry Britton
Subject: Re: Linear color space from AIM

Stephen Marsh wrote:
>Try a 1.8 or 2.2 gamma space and sharpen the gamma image the same as
>the linear image (same values in USM) - but on a duped layer. Then
>dupe the layer again so you have two USM layers. Set one to lighten
>and the other to darken. Reduce the opacity of the lighten and or
>darken layers. The results of this can then be blended in luminosity
>mode into the original data, so you have both light/dark halo
>intensity and luminosity blending.

I tried what I think you are describing, Stephen, on a still-life product shot a customer gave me recently, and the results were simply astonishing! The items on the table seemed like you could reach out and pick them up! My client was flabbergasted! (a new client - for life, now!)<g> This was a pretty flat and lifeless picture before, and besides a color correction to remove a color cast, this sharpening was all I did.

However, that does *not* mean that I did it right! So I thought I'd check to see if I did as you were suggesting, or left something out, perhaps. (Or put something in, as the case may be.)

I created a layer set in Photoshop 6 (using the Adobe 98 color space here) and placed a copy of the original (background) layer there, then sharpened it quite a lot (197 at 1.6 pixels, threshold at 3, if I remember correctly). I then made a copy of this sharpened layer within the same layer set and made the blending mode for one darken and for the other lighten as you suggested. I then set the blending mode for the layer set to luminosity. I set the opacity of the "lighten" sharpened layer to about 26%, the "darken" layer to 100%, and the opacity of the layer set to 85%. I also tried the layer set with the blending mode set to "Pass Through" with different, but still stellar results. (Pass through did cause some color shifting, while Luminosity blending resulted in slightly lighter highlights but no color shifting.) The metal parts in this composition were the biggest standout and took on a truly 3d appearance.

I also tried this on the Kodak q60 image at 307 at 1.6 pixels with threshold at 3, and set the lighten layer to 68%, the darken layer to 59%, and the layer set to 32% - again, both luminosity and pass-through gave good results, with the luminosity blend being a bit lighter. (This is a nice way to see before and after because you can simply toggle the layer set on and off to compare with the original) The text legibility was improved drastically, with no color shifts I could detect in either luminosity or pass through for the layer set. To more closely match the AIM gamma-1 example with the same sharpening settings, I had to up the layer set to 100%, blend that layer set in luminosity, and bring the lighten layer down to 22%.

Did I do this in the manner you meant, or did you also want me to luminosity blend-in a layer that was a *copy* of the sharpened AIM Gamma-1 version of the image? I tried that on the still life image, too, placing that within the same layer set as the others and set to luminosity at 28%. It did seem to bring out the texture of fabrics and tassels with more detail, and improved the contouring effect of the metal pieces. I also tried converting the entire image (with the three layers already there and a fourth added) to the AIM profile, sharpening the additional layer an equal amount as the others, then converting back to Adobe 98, again luminosity blending the resulting layer within the layer, set at 28%. This also did seem to add even more depth to the image - though the real workhorse was the darken layer, perhaps. I must experiment with both approaches more, but doing the latter additions definitely seemed to improve the 3D appearance (and was the version I showed my client.)

Thank you, everyone, for the tremendous input here - I have learned some of the most useful techniques I possess in my mere 4 weeks here already!

Terry Britton