Dan Margulis Applied Color Theory - LAB Sharpening and Other LAB Perils

Date: Sun, 12 Jan 2003 01:07:40 EST
From: Dan Margulis
Subject: LAB sharpening vs. (RGB or CMYK)/Luminosity

Folks,

For the last few months I've been experimenting with various different ways to sharpen images, a major project because there are so many strategies that need to be tested. I'm not finished yet, but I want to share one finding because it alters my previous advice.

A lot of people prefer to sharpen by attacking the L channel of LAB. I first advocated that in 1995 on the theory that this would prevent the color shifting that might happen in RGB or CMYK, and thus allow higher settings.

In some cases it's better to sharpen one or two channels of CMYK. But, I'm talking about situations where we want to do an overall sharpen.

It's been suggested for some time that doing an overall sharpen in CMYK or RGB on a luminosity layer (or reverting to luminosity after the sharpen) would give the same result as sharpening the L. As I've pointed out, this isn't true: in deep shadow areas there are technical advantages to LAB. However, I did think that there was very little difference between the two methods. So (I said, but apparently I was wrong) you shouldn't be converting out of CMYK in order to sharpen in LAB, and (I said, but apparently I was wrong) generally it isn't worth the bother to convert out of RGB to sharpen in LAB either.

I've now done a slew of images both ways, and damn if the LAB sharpen isn't significantly better a lot of the time. A lot of the time RGB/Luminosity is just as good, and I've also found one unusual image where RGB/Luminosity was better.

The reason for the superiority seems to be related to the recent thread we had about gamma. We humans are very sensitive to color variation in light areas, but not so much to variation in contrast. In darker areas, we are sensitive to contrast variation but not so much to color. Therefore, it's more important to have accurate sharpening in the darker half of the image than in the lighter half.

When we work in RGB, we're usually working at gammas of 1.8 or 2.2, the higher number indicating a darker file, with more values defining shadows and fewer defining lighter tones. The L channel of LAB, however, is at the equivalent of a much higher gamma, around 2.6 or 2.7. That has a big impact for the better if what you are sharpening is approximately a midtone in darkness, because there will be extra lightening in these areas that can be quite attractive.

Example: picture of a model with blond hair, soft background. As fleshtones are almost never as dark as a midtone and blond hair isn't either, sharpening in RGB/Luminosity and L of LAB give approximately the same results. However, change the model to a brunette and it's another story. Now the L sharpen gives a very nice sheen to the hair that isn't there in RGB/Luminosity. Similarly, in a lot of shots of relatively dark greenery the L sharpening is considerably better than (RGB or CMYK)/Luminosity, enough so that it would be worthwhile, IMHO, to convert out of either RGB or CMYK temporarily to take advantage of it.

Dan Margulis

P.S. This naturally leads to the question of whether sharpening in a 2.2 gamma RGB space, like sRGB or Adobe RGB, would be better than in a 1.8 gamma space like Apple RGB or Colormatch RGB. I've tried out a few images and my preliminary conclusion is that it's not worth worrying about.


Date: Sun, 12 Jan 2003 10:29:33 -0800 (PST)
From: Stan Schwartz
Subject: Re: LAB sharpening vs. (RGB or CMYK)/Luminosity

Dan,

When you consider all the different ways to get to sharpening, the mind boggles.

I typically have added a layer for sharpening in RGB and used Luminosity as the blending mode. Let me ask you a question about the proper sequence if I use Lab.

I just tried an image in which I converted the final RGB to Lab. I then copied the main layer to use for sharpening. I then focused on this new layer, selected just the L channel and just sharpened that channel.

I then converted back to RGB without flattening. After the conversion, I looked in detail at the image with and without the sharpening layer active. Changing this layer back and forth between Normal and Luminosity, I can see a difference, I believe.

It looks as tho' moving all the layers back to RGB removed the advantage to LAB conversion. I was hoping to be able to save the images with that sharpening layer so I don't have to recreate it for subsequent prints, which I must have in RGB.

Am I thinking wrong here?

Stan Schwartz


Date: Sun, 12 Jan 2003 20:17:52 -0500
From: Michael Demyan
Subject: RE: LAB sharpening vs. (RGB or CMYK)/Luminosity

Hi Dan:

Thank you for taking the time and making the effort to resolve this issue. Its people like you and others on this forum that will share their knowledge (and mistakes too) with those of us who want to make the best images possible, given the tools we now have. I appreciate your sharing these findings with us and always look forward to your honesty and frankness in helping all on this forum.

Mike Demyan


Date: Sun, 12 Jan 2003 21:48:39 EST
From: Dan Margulis
Subject: Re: LAB sharpening vs. (RGB or CMYK)/Luminosity

Stan writes,

>>I just tried an image in which I converted the final RGB to Lab. I then copied the main layer to use for sharpening. I then focused on this new layer, selected just the L channel and just sharpened that channel. I then converted back to RGB without flattening. After the conversion, I looked in detail at the image with and without the sharpening layer active. Changing this layer back and forth between Normal and Luminosity, I can see a difference, I believe.>>

The difference would be extremely small. This has nothing to do with the RGB vs. LAB sharpening that I was talking about.

>>It looks as tho' moving all the layers back to RGB removed the advantage to LAB conversion. I was hoping to be able to save the images with that sharpening layer so I don't have to recreate it for subsequent prints, which I must have in RGB.>>

You're comparing it to the wrong thing. If you want to see the advantage, make two copies of the original image. Sharpen one as you usually do, on a luminosity layer in RGB. Convert the other one to LAB and apply the same sharpening setting to the L channel only (no need for an extra layer). Then reconvert it to RGB and paste it as a third layer onto the original file, mode either Normal or Luminosity. Compare layer 3 to layer 2. If the picture features a lot of important dark areas, you'll probably prefer layer 3 and you can throw layer 2 away. If it doesn't, the images will still be slightly different but you probably won't have a preference.

If you think this is a worthwhile approach, then you can do just as you described initially: convert to LAB, do the L sharpen on its own layer, and then reconvert to RGB.

Dan Margulis


Date: Sun, 12 Jan 2003 20:10:53 -0800
From: Kevin Connery
Subject: LAB Curves

I just finished the 4th edition of Professional Photoshop (I already owned the first and read the 3rd), and STILL don't understand LAB curves. I can use them in straight-line fashion without any problems, but whenever I use any tweaks, kinks, or curves in them, the colors I get don't come close to what I expect.

Is there a novices guide to LAB curves somewhere? Dan's explanation goes from (for me) A to B to to M to N to O...I'm missing some key steps.

Pointers to tutorials welcomed.

Thanks!

--kdc/Kevin Connery


Date: Mon, 13 Jan 2003 01:22:27 -0600
From: "Maris V. Lidaka Sr."
Subject: Re: LAB Curves

Other than adjusting contrast using only the L channel of LAB, it is rare that you will benefit from adjusting curves using the A and B channels rather than adjusting colors in RGB or CMYK, and then only with minor tweaks in those special cases where it might help. I know of no tutorials covering adjusting A and B channel curves, other than what I have read myself in Dan's book.

Maris


Date: Mon, 13 Jan 2003 09:49:02 -0000
From: Stephen Marsh
Subject: Re: LAB Curves

Kevin Connery writes:

> I just finished the 4th edition of Professional Photoshop (I already
> owned the first and read the 3rd), and STILL don't understand LAB
> curves.

To say that LAB curves are unintuitive is probably an understatement.<g>

I tend to save them for _really_ big moves.

> I can use them in straight-line fashion without any problems,
> but whenever I use any tweaks, kinks, or curves in them, the colors I
> get don't come close to what I expect.

I often just use a duped layer in RGB/CMYK set to colour and do linear curve steepening there, although in theory LAB should give more variation if there is a lot of subtle hues of similar values - although you may really have to look to tell.

As for bending the curve...

I would start by placing a control point at the 50% (neutral point in AB) AB curve point, so that existing neutrals are not shot.

Next I would place fixed colour samplers in the key areas that are being affected and some in areas that should not be affected by the edits. If working for a CMYK output, the expected CMYK reconversion values and gamut warning indicators would be viewed when editing to add to the visual evaluation of the monitor. Most moves would only be minor, as in 2-5 points movement - but it all depends on what you are doing.

> Is there a novices guide to LAB curves somewhere? Dan's explanation goes
> from (for me) A to B to to M to N to O...I'm missing some key steps.

There is a curve editing technique known as 'lock down curves' - which is not very good for most RGB/CMYK edits, but it may work better for the AB of LAB with more minor edits. A control point as placed at each of the 10, 20, 30 > 90% points - which locks down the curve through the tonal range. You can then choose to may 'incremental' edits to sub tonal ranges (say between 80-90%). Or delete a range of control points to have more control but still retain original values in other places of the curve.

LAB curves are great, but for most tasks I wimp out and use colour and luminosity blend curves in RGB/CMYK - as the finer incremental control and different handling of neutral tones is often easier than in LAB. LAB does have many advantages and possibilities, however blend modes often reduce the need for LAB for me (but blends in RGB/CMYK are not the same thing, sometimes similar but not the same and it's those little differences that LAB curves can produce which make it worth using, even if it is just to see how what the difference is over a colour blend edit in RGB/CMYK).

As for other sources of practical info on LAB, not to mention LAB curves - most other authors and 'names' shun the use of LAB, so the links below all seem to be influenced by Dan.<g>

http://members.aol.com/clehan/tips//lab.color.pdf

https://www.ledet.com/margulis/LABCorrection.pdf

http://ep.pennnet.com/Articles/ArticleDisplay.cfm?Section=Archives&Subsection=Display&ARTICLE_ID=92230&KEYWORD=dan%20margulis&x=y

http://ep.pennnet.com/Articles/ArticleDisplay.cfm?Section=Archives&Subsection=Display&ARTICLE_ID=76110&KEYWORD=dan%20margulis

https://www.ledet.com/margulis/ACT_postings/ACT-LABLoss.html

http://www.varis.com/ColorTheorySuccess2/index.htm

http://www.varis.com/ColorTheorySuccess/index.htm

http://www.dbphoto.net/moire

Regards,

Stephen Marsh.


Date: Mon, 13 Jan 2003 10:28:17 -0000
From: Stephen Marsh
Subject: Re: LAB sharpening vs. (RGB or CMYK)/Luminosity

Stan writes:

> When you consider all the different ways to get to
> sharpening, the mind boggles.

Mine is past the 'boggle' point.<g>

> I typically have added a layer for sharpening in RGB
> and used Luminosity as the blending mode.

A very good method - why sharpen the colour component of the image, which is useless for adding the appearance of sharpness and just may amplify artifacts or colour noise?

> I just tried an image in which I converted the final
> RGB to Lab. I then copied the main layer to use for
> sharpening. I then focused on this new layer, selected
> just the L channel and just sharpened that channel.

You could just use the one layer, unless you were adding layer masks or blend-if sliders into the mix - do you really need the file to be twice as big just for the sharpen?

> I then converted back to RGB without flattening. After
> the conversion, I looked in detail at the image with
> and without the sharpening layer active. Changing this
> layer back and forth between Normal and Luminosity, I
> can see a difference, I believe.

As Dan indicated, this is not the way to do the evaluation, although if you do see differences then I would say that they have taken place due to the mode conversion.

What you probably want to do is dupe the RGB layer in question and do not change modes, simply sharpen that duped layer in RGB in normal mode. Then toggle between normal and luminosity to see if there is a difference.

Then if you want to compare the RGB sharpen to LAB, do the sharpen in RGB and fade to luminosity or set the duped sharpen layer to luminosity...and perform the same sharpen settings on an unsharpened copy that is in LAB (only sharpen L).

Then this L of LAB sharpen is compared to the RGB luminance simulation sharpen.

Hope this helps,

Stephen Marsh.


Date: Mon, 13 Jan 2003 11:57:26 EST
From: Dan Margulis
Subject: Re: LAB Sharpening vs. (RGB or CMYK)/Luminosity

The following came to me privately this morning, but it certainly seems appropriate to forward it to the list.

Dan Margulis

Subj: LAB Sharpening vs (RGB or CMYK)/Luminosity
Date: Monday, January 13, 2003 10:23:18 AM
To: DMargulis

Dan,

In your post over the weekend to the colortheory list you were discussing sharpening in LAB vs RGB or CMYK/Luminosity.

I work in Photoshop on an everyday basis and am sharpening images some most of the time. My work is primarily involving people (portrait, wedding and bar mitzvah work). I use your edge sharpening technique with great results. I occasionally use Nik's Sharpener Pro as well as some other techniques. However, I've never converted to LAB to do my sharpening.

It sounds to me like what you are proposing is that, especially with low key images (most of my work), we should convert the file from RGB to LAB, select the L* channel and do our sharpening there and then convert back to RGB or CMYK for our output. Almost 100% of my work is output in RGB. Am I on the right track here or am I missing something?

Thank you for your time and attention.


Date: Mon, 13 Jan 2003 12:19:10 -0500
From: Dolores Kaufman
Subject: Re: LAB sharpening vs. (RGB or CMYK)/Luminosity

Dan,

I just want you to know how much you are appreciated. It is SO refreshing to find someone with so little ego investment in their findings. You embody the true spirit of scientific inquiry that, unfortunately, few scientists themselves seem able to muster. I'm sure I speak for many in saying Thank You for your candor and your generosity. You have given all of us so very much.

Dolores Kaufman


Date: Mon, 13 Jan 2003 18:46:46 EST
From: Dan Margulis
Subject: Re: LAB Sharpening vs (RGB or CMYK)/Luminosity

A list member writes,

>It sounds to me like what you are proposing is that, especially with low
>key images (most of my work), we should convert the file from RGB to LAB,
>select the L* channel and do our sharpening there and then convert back
>to RGB or CMYK for our output. Almost 100% of my work is output in RGB.
>Am I on the right track here or am I missing something?

Close. What I'm saying is that doing that with a low-key image should give you better results than doing the equivalent amount of sharpening in RGB, whether or not you fade to luminosity afterward.

What I'm *not* saying is that this is absolutely the best way to sharpen a low-key image. I think I have better suggestions but I'm not quite ready to post them. But I've compared enough images now to be sure that sharpening the L is often better than sharpening RGB/Luminosity, enough better that it could be worth a workflow change for certain people. So, I wanted to get that information out now. For the other stuff, stay tuned.

Dan Margulis


Date: Mon, 13 Jan 2003 21:17:52 EST
From: Dan Margulis
Subject: Re: LAB Curves

Kevin writes,

>Is there a novices guide to LAB curves somewhere? Dan's explanation goes
>from (for me) A to B to to M to N to O...I'm missing some key steps.

Straight-line AB curving is easy enough for anybody. More complex AB curving is extremely powerful but it takes a *lot* of practice and is real hard to learn without supervision. Of the people who've taken my introductory course I'd guess fewer than a dozen were fully comfortable with AB curves when they left, but many more have gotten good at it with later practice.

It's doubtful that any written materials would suffice to make one an AB curving maven absent a lot of investment of time. Nevertheless, I'm intending a three-part series of articles on the many uses of LAB later this year and will cover advanced curving to some extent.

There are four full chapters on LAB or LAB-like maneuvers in Professional Photoshop and that's about all the room I think I can justify. There are so many things that LAB does well that I have toyed with the idea of writing a book about LAB only, with the idea of making it more accessible to the masses. But, I just don't think there's a big enough market to make it worthwhile.

At Photoshop World in LA in February, I'm going to be speaking for two hours all about LAB. That lecture is up against some very stiff competition from other Photoshop gurus so I'm not expecting people to be breaking down the doors to get in to hear me. If it turns out that there's some unusual interest, then I might revisit the idea of a book.

Dan Margulis


Date: Tue, 14 Jan 2003 02:03:43 -0000
From: Stephen Marsh
Subject: Re: LAB Sharpening vs. (RGB or CMYK)/Luminosity

> I work in Photoshop on an everyday basis and am sharpening images some most
> of the time. My work is primarily involving people (portrait, wedding and bar
> mitzvah work). I use your edge sharpening technique with great results.

I have never heard Dan mention the edge isolation technique - although I am sure he is aware of it. One thing that Dan does recommend for this type of work is targeted channel sharpening - CK for skintones in portraits, or R in RGB using a luminosity blend layer.

You can also load the inverted R/C channel data as a selection to mask the skintones while still hitting the hair and eyes etc (use the channel data as is or dupe and enhance it so that it is more like a black and white mask channel). This will add some more snap over targeted channel sharpening...or combine both methods.

Regards,

Stephen Marsh


Date: Mon, 13 Jan 2003 18:44:17 -0800 (PST)
From: Stan Schwartz
Subject: Re: Re: LAB sharpening vs. (RGB or CMYK)/Luminosity

Thanks. I think I didn't articulate my observation clearly. What I meant was this:

After I had done the sharpening in LAB and converted back to RGB, I tested the sharpening process by toggling the sharpened layer between Normal and Luminosity. I did this at 100% (View Pixels) to see if I could see any color shift in the sharpened edges.

If the sharpening retained the advantages of sharpening in L, I reasoned that even after the conversion back to RGB, only the luminosity of the sharpened pixels should have been affected.

Therefore, I further reasoned that if only luminosity were affected, I should see no difference when I toggled between Normal and Luminosity. But I did see some color shifts.

So I wasn't doing this to compare sharpening methods but just to see if luminosity-only sharpening "survived" the transition back to RGB.

I hope I am being as clear as I am verbose.

Stan


Date: Mon, 13 Jan 2003 21:17:48 -0600
From: "Maris V. Lidaka Sr."
Subject: Re: LAB Curves

Dan Margulis wrote:

> It's doubtful that any written materials would suffice to make one an
> AB curving maven absent a lot of investment of time. Nevertheless,
> I'm intending a three-part series of articles on the many uses of LAB
> later this year and will cover advanced curving to some extent.

Is that a promise?<g>I wouldn't miss those for the world.

> At Photoshop World in LA in February, I'm going to be speaking for
> two hours all about LAB. That lecture is up against some very stiff
> competition from other Photoshop gurus so I'm not expecting people to
> be breaking down the doors to get in to hear me. If it turns out that
> there's some unusual interest, then I might revisit the idea of a
> book.

I would be there, but have to travel to Hawaii on February 17th, 2 days
before opening (for business - really). Are you preparing any written
materials for the lectures, for distribution for those who can attend or for
sale to those who cannot?

Maris


Date: Tue, 14 Jan 2003 13:57:59 -0000
From: Stephen Marsh
Subject: Re: LAB sharpening vs. (RGB or CMYK)/Luminosity

Stan S wrote:

> After I had done the sharpening in LAB and converted
> back to RGB, I tested the sharpening process by
> toggling the sharpened layer between Normal and
> Luminosity. I did this at 100% (View Pixels) to see if
> I could see any color shift in the sharpened edges.

Hi Stan, close - but this extra variable is throwing things off. Forget the luminosity blend toggle on the RGB/LAB/RGB image that had sharpening performed in L of LAB.

What you want to do is compare the L sharpen to the same sharpen in RGB done in normal mode (which affects both colour and tone).

THEN compare the L sharpen to the RGB sharpen with the luminance fade or blend to see the difference between LAB and the simulation in other modes.

L of LAB does isolate the tone from colour, no doubt about it. There is no real need to luminance blend and toggle that layer when using this in production - although this will help preserve the original files colour component from the LAB trip if there was a change in colour from RGB/LAB/RGB. I will often dupe and go to LAB and then reblend in either luminosity or colour blend modes to isolate the LAB files contribution to the edit in question - but this is a production move for a specific reason and not a toggle test as in your case.

The step you were doing is showing you the difference between a layer in normal mode and that same layer being blended into underlying data via a luminance simulation (approx 30r59g11b). As previously noted, the RGB/CMYK simulation of luminance is close to L of LAB in some respects - but it is not LAB. Colour is just divorced from tone in the blend, since the file is still in RGB/CMYK it will not behave as LAB or other colour modes that truly separate colour from tone.

> If the sharpening retained the advantages of
> sharpening in L, I reasoned that even after the
> conversion back to RGB, only the luminosity of the
> sharpened pixels should have been affected.

Affected by the sharpen, then yes I agree. All pixels have been affected in the file though, as there was a mode change. Thus the blending of luminance/colour as described above to limit this if it is a concern.

> Therefore, I further reasoned that if only luminosity
> were affected, I should see no difference when I
> toggled between Normal and Luminosity. But I did see
> some color shifts.

As mentioned above, you are seeing the difference of a layer in normal mode and that same layer being luminosity blended into underlying data - which is not a valid test of RGB or CMYK luminance sharpening vs. LAB.

> So I wasn't doing this to compare sharpening methods
> but just to see if luminosity-only sharpening
> "survived" the transition back to RGB.

It does - trust me.<g>An easier test is tonal moves or dodge/burn or other larger contrast shifts. Or comparing individual channel noise which should be lesser on the RGB/LAB/RGB file than the one that had normal mode RGB/CMYK sharpening (use strong USM to evaluate).

Just don't introduce the luminance blend toggle into things. Only compare the regular edit in RGB/CMYK in normal mode to the same one done in LAB. Then as a further step see how the luminance simulation of RGB/CMYK compares.

Hope this makes sense.

Stephen Marsh.


Date: Tue, 14 Jan 2003 08:02:01 -0600
From: Ron Bean
Subject: Re: LAB Curves

Dan Margulis writes:

>Straight-line AB curving is easy enough for anybody. More complex AB curving
>is extremely powerful but it takes a *lot* of practice and is real hard to
>learn without supervision.

Is this partly because the LAB colorspace is so big that the
usable part is compressed into the center of the curve?

Would it be easier if we had a LAB colorspace whose gamut was
closer to the other colorspaces? If we can have a wide-gamut
CMYK, why not a smaller-gamut LAB?

I don't know if it's even possible to make an ICC profile for
LAB. But if not, maybe Adobe could include a smaller-gamut LAB as
an option in some future version of Photoshop.

>There are four full chapters on LAB or LAB-like maneuvers in Professional
>Photoshop and that's about all the room I think I can justify. There are so
>many things that LAB does well that I have toyed with the idea of writing a
>book about LAB only, with the idea of making it more accessible to the
>masses. But, I just don't think there's a big enough market to make it
>worthwhile.

How about including it on the CD in some future edition if PP.
Or sell it separately as some kind of "ebook".

I wonder if it would be practical to put *all* of the images on
the CD and just print the book in B/W...


Date: Tue, 14 Jan 2003 07:26:36 -0700
From: Andrew Rodney
Subject: Re: LAB Curves

Ron Bean wrote:

> Would it be easier if we had a LAB colorspace whose gamut was
> closer to the other colorspaces? If we can have a wide-gamut
> CMYK, why not a smaller-gamut LAB?
> I don't know if it's even possible to make an ICC profile for
> LAB. But if not, maybe Adobe could include a smaller-gamut LAB as
> an option in some future version of Photoshop.

Don1t think that1s possible. LAB is self defining so no profile is needed (unless there are still to color temps to worry about; D50 and D65). It1s a device independent colorspace that has no relationship to any physical device (input, display or output). One disadvantage of LAB is it1s huge size!

Andrew Rodney


Date: Tue, 14 Jan 2003 10:33:49 EST
From: Dan Margulis
Subject: Re: LAB Curves

Maris writes,

>I would be there, but have to travel to Hawaii on February 17th, 2 days
>before opening (for business - really). Are you preparing any written
>materials for the lectures, for distribution for those who can attend or
>for sale to those who cannot?

I have handouts for all my sessions. However, 1) they're entirely B/W; 2) they are intended to supplement the lecture and would be difficult to follow for somebody who didn't hear it; 3) at least in the case of the LAB session, the handout is somewhat of a stalking horse for some intended columns later on.

So, I do not intend to make the handouts available except to those who attend. The columns, when they come, will be in color and a lot more understandable.

Dan Margulis


Date: Tue, 14 Jan 2003 10:35:41 EST
From: Dan Margulis
Subject: from an ex-pupil-sharpening, etc.

********FORWARDED MESSAGE********

Dan,

My name is Al Ferreira - I'm a photographer who took your course in NJ back in 2001.

I follow the daily exchanges in the color theory list, and much appreciate your input and those of many others. Due to the expert level of many of the participants, I prefer to just lurk on the sidelines.

However- your recent post about sharpening images reminded me of how I sharpened files after I attended your seminar and was mostly unhappy with the results. The reason is that I believe that there is a difference between sharpening original digital camera files and scanned files from film. I've deducted (correct me here if I'm wrong) that many of your sharpening suggestions introduce a "jagged" look to digital camera files making them look obviously different from film (scanned) files and in my opinion, inferior.

As a photographer who's been shooting for 35 years - I love the "look" of film and try to manipulate and correct digital camera files to achieve similar color and "feel" characteristics. It's been my experience that "edge" sharpening introduces that undesirable "jagged" look, wether it was achieved thru sharpening the black or other channels in cmyk or simply using a higher threshold level that affects edges primarily. I usually sharpen in LAB. I sense that a lot of the digital sharpening discussions are not intended for digital images. I've used at times a threshold of zero, sharpening every pixel overall, often it gives a more film look.

Simply put - I believe that much of the sharpening techniques have evolved from scans and not original digital camera files, and have to be updated to reflect the real world ( yes - I know how you feel about photographers :-) )

Your are welcome to take this dicussion publicly on the forum, I was just afraid to stick my head out there.

Regards,

Al Ferreira
Al Ferreira Photography, Ltd.
www.afpltd.com


Date: Tue, 14 Jan 2003 08:52:31 -0700
From: Andrew Rodney
Subject: Re: from an ex-pupil-sharpening, etc.

Dan Margulis wrote:

> The reason is that I believe that
> there is a difference between sharpening original
> digital camera files and scanned files from film.

There is a considerable difference between the two and the methods of sharpening digital capture need to be taken into account differently than a scanned image (which is filled with grain and quite different noise).

Andrew Rodney


Date: Tue, 14 Jan 2003 10:03:16 -0600
From: "Susan and John Opitz"
Subject: L*A*B curves

Doing lab moves for casts are(for me) a lot better than just sampling the color(cast) and making a new layer,changing the layer to color and inverting it,adjusting the opacity of the layer. I feel this is the punie girly man way of doing this. Using the "a" or "b" channel gives a better look to the image. It's not as flat looking(color) like when you invert that color cast color. The "a',"b" can be used to pump up other colors in the image as well.

Professional Photoshop(new version) has some examples on how to use the "a","b" curves. The curves are for the images in the examples. But, the
curves(technique) can be used for other purposes as well.

John Opitz


Date: Tue, 14 Jan 2003 11:05:39 -0500
From: Floyd Rominski
Subject: in appreciation

Dan,

Your Feb. lecture would be impossible for me to attend :( , but as mentioned by Maris L. i would get in line to purchase a pdf. of same.

Dolores K.s post . . .articulate and to the point, expresses what most of us experience (certainly true for those who have participated in your classes).

Thanks so much , you jolly genius.

Regards,
floyd rominski


Date: Tue, 14 Jan 2003 09:58:34 -0700
From: Ron Kelly
Subject: Re: from an ex-pupil-sharpening, etc.

Andrew Rodney wrote:

> There is a considerable difference between the two and the methods of
> sharpening digital capture need to be taken into account differently than a
> scanned image (which is filled with grain and quite different noise).

Yes, this is no doubt the case. What's interesting to note is that now we are at another point in our evolution where we have to consider: what is desireable?

Traditional film, scanned and output, must be sharpened in the process because the scanning introduces fuzziness. Exactly why that happens on the very best equipment, say, a top-end drum scanner, I don't know. But it does, and if it isn't sharpened it looks bad.

In any case we now have to re-evaluate our aesthetic.

Some of us photographers used to look at a noisy scan and think that it was good (and sharp) because "you could see the grain!"Back then, we were divorced from the digitizing process, and could not always know why some scans looked better than others, or indeed what was a good scan. Now we may know better, and the look of grain is not necessarily as good as it once was.

We've been looking at film so long it has become the standard. How long will it be until we can't stand the look of film grain, and find it just as repulsive as bad sharpening? Should we be able to see any texture at all that isn't attributable to the subject?

Cheers,
Ron Kelly


Date: Tue, 14 Jan 2003 10:15:10 -0700
From: Andrew Rodney
Subject: Re: from an ex-pupil-sharpening, etc.

Ron Kelly wrote:

> Traditional film, scanned and output, must be sharpened in the process
> because the scanning introduces fuzziness. Exactly why that happens on
> the very best equipment, say, a top-end drum scanner, I don't know.

Simple. ALL capture devices need some sharpening (assuming the device isn1t doing this as part of the input process). The world we live and see in is continuous tone (truly contone). When you take something and digitize it, you produce nothing more than a mosaic of (for now) square pixels. That softens things quite a bit! So some kind of sharpening is necessary. Since we are dealing with pixels, the actual resolution of the file being sharpened is a factor. In addition, the output device as well as the subject plays a huge role. For this reason, it is usually a good idea to do a double dose of sharpening (what we like to call pre-sharpening and post (output) sharpening. Pre-Sharpening should be very minor and only enough to compensate for the digitizing process. Archive that and then sharpen from there once the final size, and output requirements have been made. This is not difficult to do but adds an extra step in the process (unless you live comfortably with the scan once, use once workflow).

We (Pixel Genius) who have some automated sharpening routines in PhotoKit are working on some more complex sharpening routines that will take input device (digital camera verses scanner) into account and we will have different file sizes and different camera settings (3mp, 5mp etc) available to do both pre and post sharpening. All the work would be done on individual layers so users can further tweak the effect using apply modes and opacity and keep the underlying data intact which is always a good idea. There is no question that data from a digital camera is vastly different in how it needs to be handled than a scanned image. Protecting certain areas from increased noise due to sharpening is quite important and each file has difference where the sharpening needs to be applied or not. Banding in smooth areas of a digital camera capture are far more apparent than what you1d see even on the best scanner due to the grain structure of any film that simply isn1t in a file from a digital camera. Adding noise may be necessary from some (certainly when you have to composite a scanned image onto an image shoot on a digital camera).

Andrew Rodney


Date: Tue, 14 Jan 2003 09:30:59 -0800
From: "Seth Campbell"
Subject: on the topic of sharpening

To Dan and the others, On the topic of sharpening, I think that it was last year or maybe the year before in Napp's Photoshop User magazine 101 tips there was a tip to use a copied layer > Highpass filter> overlay or softlight blend mode to sharpen a image. I have used this a lot on fine textured images. When I set the setting for the highpass layer I set it so that I can see allot of detail and then I back it off. I was wondering what you thought about this method? Is there something that I should look out for.

Thanks,

Seth Campbell

Smith/Walker Design
990 Industry drive
Seattle, WA 98188


Date: Wed, 15 Jan 2003 01:49:49 -0000
From: Horacio Peña
Subject: Re: on the topic of sharpening

Seth Campbell wrote:

> On the topic of sharpening, I think that it was last year or maybe the
> year before in Napp's Photoshop User magazine 101 tips there was a tip to use
> a copied layer > Highpass filter> overlay or softlight blend mode to
> sharpen a image....

I have been using something similar for dark images, using Highpass filter to both lighten and sharpen with one curve.

1- Duplicate layer
2- apply Highpass (radius 10)
3- AutoLevels to increase contrast of "highpassed" layer
4- Ctrl-Alt-~ (or Crtl-click on RGB channel in the channels dialog) to create a selection out of the luminance of layer
5- Ctrl-H to hide those nasty ants :)
6- change to the background layer
7- open curve dialog and start pushing the curve upwards with the most ridiculous curve ypu can draw and whatch ... more details and more sharpening whith just one curve !

All these can be easily converted to an Action.

This is my $.01 tip...

Best regards
Horacio


Date: Wed, 15 Jan 2003 13:52:19 -0800
From: "Stephen Marsh"
Subject: Re: Advanced USM Options (was - Linear color space from AIM)

Terry writes:

> 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!

Hi Terry, a product shot...that is a very suitable subject for backing off on the light halo (portraits too...the more you look the less you need full light halo intensity). But that's just a guess, not knowing the product...but knowing how many different types of product shots behave, I have a good idea why this worked so well, or at least well in certain places. You may want to explore targeting different image frequencies with different amounts of USM (finer detail with small fine USM and gross detail with thicker harder USM, as well as the variable halo intensity).

What I find astonishing is that Adobe do not give their 'power' users an advanced radio button tab or even an optional install USM filter. As many users find the relationship between the existing three sliders too complex, a beefed up USM may not be for all users. For many years scanner users have had the ability to control light and dark halo intensity and other variables beyond the three defaults in the Adobe USM. As input applied sharpening has fallen out of favour in many of today's workflows - it makes sense to have the powerful USM options in Photoshop as many do not use them in scanning anymore. Of course you can do many tricks in Photoshop to work around the basic unsharp mask sharpening filter limitations, but that is not the point. These hacks often make the file twice the file size or more and are poor in interactive use - everything needs to be accessed from the one dialog with a live preview.

> 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.

So you did not add any RGB or CMYK luminosity curves or even better a LAB tweak of luminosity? One of the best ways to 'sharpen' is not to sharpen, but to add contrast via global curving or other more selective methods. Once there is a good tonal range (if suitable, perhaps this product shot is flat for a reason such as content and there is no way to get realistic full range).

Or do you mean that the colour caste correction you did was in normal blend mode, so you were affecting both colour and tone in the correction - although you may have only been concerned with the colour and just let the luminosity fall naturally with the colour edit? Only asking as I am not sure exactly what you mean - if I am asked to knock out a colour caste then I would not change the brightness (so I would use colour blends/fades or AB of LAB).

The AIM demos seem to be stacked to favour hi dark halo and low light halo sharpening, which is what an extreme gamma 1 sharpen does - it places way too much emphasis on the dark halo and really backs off on the light. The variable halo intensity sharpening methods done in regular gamma or LAB or CMYK show that you can have a better sharpen than standard USM or gamma 1 USM, without changing to gamma 1 and having more control and less visible and invisible damage to the file than in gamma 1 sharpening.

> 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.)

Some basic ways to produce variable USM halo intensity:

i) Two USM layers, one set to lighten the other darken - play with opacity in the lighten or perhaps darken layers. The steps would be dupe bg layer, apply USM and fade to luminosity, then set to darken blend mode, dupe this again and set to lighten and play with opacity etc. Con - three layers.

ii) Use a single duped layer and use the layer option blend if sliders to back off on the light halo (opt/alt click the white slider and split the left half over to the extreme left shadow point or lesser). Again, this is done on a luminosity fade normal blend sharpen layer. Pro - two layers, plus you can see the split intensity effect while you sharpen all in one go.

iii) Use LAB mode to sharpen and use layer blend if as in step ii. Pro - two layers.

iv) History or pattern fills. The source data is a luminosity sharpen. A lighten fill is used at XX% and a darken fill is used at 100% or XX%. Pro - two layers.

There are other ways to do all this, as in doing this on a duped file and then merging all three layers (lighten XX%, darken 100% and normal bg layer) into one layer and then blending this in luminosity over the original bg layer and then introducing other options such as advanced blending where you ignore certai channels from the mix.

As for your blending luminance blending questions, the methods above cover that.

a note on layer sets, be wary. You may have a 50% lighten and 100% darken USM layer set - but changing from pass through to luminosity or merging the layer set ignores the effect of the 50% layer and treats it as 100%...and who knows what else gets ignored. If I want the layer to blend at 50% that is the effect I want - not 100%!<g>The actual layer still is set to 50%, but it is being applied at 100%. My understanding of layer sets and blending must be off, or this is a bug or an application limitation (like layer masks disabling the fade command).

So thus the above methods which avoid layer set opacity bug issues.

Regards,

Stephen Marsh.


Date: Wed, 15 Jan 2003 15:06:19 -0800
From: "Stephen Marsh"
Subject: Re: High Pass (was - on the topic of sharpening)

Seth Campbell writes:

> On the topic of sharpening, I think that it was last year or maybe the
> year before in Napp's Photoshop User magazine 101 tips there was a tip to use
> a copied layer > Highpass filter> overlay or softlight blend mode to
> sharpen a image. I have used this a lot on fine textured images.

Hi Seth, I have many thoughts on sharpening, and high pass filtering has it's place among them - just as unsharp mask and other methods have their place in the process.

I have to give credit to the high pass introduction to an earlier source than NAP, but it does not really matter who get's credit as this is ancient info in the image processing realm - it is just that for most Photoshop users high pass filtering is a fairly new thing. Audio processing and image processing are quite similar in some ways. I was blown away the first time I played with audio low/high pass filtering (I should have chosen audio and not prepress as a career).

> Is there something that I should look out for.

Noise amplification, excessive edge halo width (too high high pass settings). Just some of the regular things as with USM but as you have less control with high pass you have to take more care.

I have been doing a bit of research into basic image processing over recent months, which has been a great way to waste time - although it has given me a better understanding of many of the operations Photoshop performs behind the scenes. I have made my own find edges filters/actions that just about 98% matches the results of the native find edges filter and made a few other filter kernels/actions. Big deal! This is entry level graphics geek stuff which is beyond my wants/needs but as an education experience it has been worth it. As Dan says in his book, you don't have to know the mechanics of USM to use it (but that did not stop me from wanting to know more<g>).

The basic operation of USM is done by subtracting a blurred original from the original. The image is made sharper by removing low frequency data from the image so you can see the high frequency data. So one way to think of USM is that it removes blur so you can see the existing sharpness of the image. On the other hand - High Pass removes low frequency data and only passes through higher frequencies (second derivative or Laplacian). When we sharpen via high pass methods we are adding existing sharpness to the image.

Both methods amplify local edge contrast - in similar but different ways. USM sharpens with built in noise suppression, then when you factor in thresholding as well there is added noise control. High Pass seeks out all the noise and edges in the image and then you add them back to the original. So this is why a high pass and USM of the same approx overall effect will always deliver a very slightly sharper image via high pass.

There are some basic steps than can be done when high pass sharpening:

* Prefilter with deskeckle or minor g/blur before high passing (less sharp but less noisy)
* High Pass and then despeckle or minor g/blur (more sharp than above but less noisy than using no noise suppression)
* Or combine both of the above methods of pre & post smoothing using smaller
amounts via fades
* Desaturate the high pass blend layer after high pass filtering
* Introduce luminosity blends (see my other recent post for more on this subject)
* Introduce variable halo intensity (see my other recent post for more on this subject)
* Blend modes from soft light, overlay, hard light, linear light or perhaps vivid light
* Various opacity settings
* Try smaller high pass values to start and watch for halo issues as you increase size

So if noise is a concern - then USM or other methods may be better. If you don't have to worry about noise or are after a little more pop - then high pass methods might be better, although the halo is different.

Then there are other filtering methods, such as laplacian of gaussian [LoG] or band pass filtering (difference of gaussians [DoG] or difference of medians, difference of box/mean etc). High pass and other methods can also be performed via adobe or third party custom filter convolution or in some cases layer or apply image blends or layers and filterings.

Sharpening can be made as simple or complex as you like.

P.S. Image processing gurus and tech heads - sorry for any errors in the above, I am new to the 'back end' process of filtering and I am not a programmer. I am happy to be further educated if the above offends.<g>

Stephen Marsh.


Date: Tue, 14 Jan 2003 23:44:20 EST
From: Dan Margulis
Subject: Re: LAB Curves

Ron writes,

>Is this partly because the LAB colorspace is so big that the
>valuable part is compressed into the center of the curve?

That's a big part of it. But mostly it's that the concept is a foreign one. You need to visualize how the A channel, which is green-to-magenta, behaves in an object that's orange and blue. Very difficult to do that until you get used to it. Consider how many RGB users think that learning CMYK is very difficult (and vice versa) when in fact the two are extremely close to one another. Now, consider how difficult these people find learning the language of AB, which really is radically different.

>Would it be easier if we had a LAB colorspace whose gamut was
>closer to the other colorspaces? If we can have a wide-gamut
>CMYK, why not a smaller-gamut LAB?

Yes. That would make life easier. Linotype-Hell actually implemented something like it in their products, a different LAB called LAB (LH) which emphasized the printable areas.

>Maybe Adobe could include a smaller-gamut LAB as
>an option in some future version of Photoshop.

Adobe has not distinguished itself by its understanding of what LAB is and does. I suspect you would find them too busy designing histogram palettes and ways to inconvenience users of non-Adobe products to be interested in something so practical. As Stephen has pointed out, the USM implementation in Photoshop may have been fine for version 2.0 but it leaves a lot to be desired in today's world. If they can't find the time to fix that, it's hard to imagine they'd want to mess with a new LAB definition.

Dan Margulis


Date: Wed, 15 Jan 2003 00:20:51 EST
From: Dan Margulis
Subject: Re: from an ex-pupil-sharpening, etc.

Al writes,

>>I've deduced (correct me here if I'm wrong) that many of your sharpening suggestions introduce a "jagged" look to digital camera files making them look obviously different from film (scanned) files and in my opinion, inferior.>>

If the files look jagged, then they're oversharpened, no matter what their origin.

A digital camera is nothing more than a scanner with a variable focal length, so there's not going to be all *that* much difference between digicaptures and scanned film. There are imperfections in CCDs that can cause fringing or weird colors. As against that, film has noise and scanning film introduces more imperfections.

There are going to be some differences, that's for sure. In files of the same resolution, digital captures will start with better edges than scanned film. Sometimes that helps sharpening, for obvious reasons, but sometimes it hurts. For example, in small areas that you really don't want to emphasize there is often enough crisp detail in digicaptures that you can't exclude it from the sharpen, whereas in film scans it's just noise that can be controlled by the threshold setting.

But the bottom line is, it's the character of the image, not the way it was acquired.

>>As a photographer who's been shooting for 35 years- I love the "look" of film and try to manipulate and correct digital camera files to achieve similar color and "feel" characteristics...I've used at times a threshold of zero, sharpening every pixel overall, often it gives a more film look.>>

That's fine, but it's a personal preference.

>>Simply put - I believe that much of the sharpening techniques have evolved from scans and not original digital camera files, and have to be updated to reflect the real world.>>

Almost all of the fancy sharpening techniques being discussed in this thread date from the age of digicams, not before. We now have lots of new sharpening options that really weren't feasible when computers weren't as fast as they are today.

Dan Margulis


Date: Tue, 14 Jan 2003 23:52:17 -0600
From: Ron Bean
Subject: Re: LAB Curves

Dan Margulis writes:

>>Maybe Adobe could include a smaller-gamut LAB as
>>an option in some future version of Photoshop.

>Adobe has not distinguished itself by its understanding of what LAB is and
>does...

>... it's hard
>to imagine they'd want to mess with a new LAB definition.

Maybe they could just scale the dialog box so it's easier
to draw the curves...


Date: Wed, 15 Jan 2003 07:08:06 -0700
From: Andrew Rodney
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Dan Margulis wrote:

> A digital camera is nothing more than a scanner with a variable focal length,
> so there's not going to be all *that* much difference between digicaptures
> and scanned film.

Should I send you a CD of the same 4x5 or medium format back using both film and either a scan back (best) or a 1 and 3 shot capture? Not anything alike as far as the files. The mechanics may be similar. The files are night and day.

Even with neg film (who can scan that???<g>), you1ve got maybe 7 stops of range and with a digital camera, up to 13 stops. In digital capture, you have NO grain, no H&S Curve and one less set of optics to get 11s and zero1s (the scanner). You are either over simplifying the case or you need some files to look at (and Dan, I1m not talking about your consumer Nikon toy camera although that does produce nice files that in many ways look nothing like film).

> There are imperfections in CCDs that can cause fringing or
> weird colors.

Very little if any when you work with quality equipment. The Betterlight scan back is a 3scanner on a stick2 but with no film, it1s producing vastly superior (and different files) that no scanner on this planet can compete with for pure quality of file. The files need to be handled differently because they are so different.

> But the bottom line is, it's the character of the image, not the way it was
> acquired.

Hogwash.

Andrew Rodney


Date: Wed, 15 Jan 2003 07:02:06 -0700
From: Andrew Rodney
Subject: Re: LAB Curves

Dan Margulis wrote:

> Adobe has not distinguished itself by its understanding of what LAB is and
> does.

Hold the presses, Dan and I agree! I1ve been asking Adobe to implement LCH
corrections ala LinoColor and NewColor for years. LCH is VERY easy to learn
and very powerful. It makes using LAB a cinch.

> I suspect you would find them too busy designing histogram palettes and
> ways to inconvenience users of non-Adobe products to be interested in
> something so practical.

Like 16 bit editing, who would use it in the installed based of PS users? A rhetorical question of course.

> As Stephen has pointed out, the USM implementation in
> Photoshop may have been fine for version 2.0 but it leaves a lot to be
> desired in today's world.

Again I agree that the old USM tool needs major revamping and more power. I can do more stuff I my Imacon scanner software which is pretty shocking!

Andrew Rodney


Date: Wed, 15 Jan 2003 11:34:22 EST
From: Dan Margulis
Subject: Re: on the topic of sharpening

Seth writes,

>>I think that it was last year or maybe the year before in Napp's Photoshop User magazine 101 tips there was a tip to use a copied layer > Highpass filter> overlay or softlight blend mode to sharpen a image. I have used this allot on fine textured images. When I set the setting for the highpass layer I set it so that I can see allot of detail and then I back it off. I was wondering what you thought about this method? Is there something that I should look out for.>>

This is another style of sharpening that I was going to write about later, but here goes. It gets quite a different look than conventional USM. However, as described above is a fairly inflexible way of getting that look.

The idea of any kind of sharpening is to emphasize certain objects and transitions. Conventional sharpening does this by creating contrasting edges, or halos, around them.

The edges are intended to be too subtle for casual viewers to perceive, but they're easy enough to see if you blow up the image on screen or look at it under a loupe in print.

A different approach is to create the needed contrast by grossly larger halos. To do this, obviously, the halos have to be almost imperceptibly different from the object, because they cover so large an area. If it's done right, however, it can add what seems to be an extra dimension to the picture.

Achieving this with the high pass filter as described above works, but we can get substantially the same result by using the USM filter with an extremely high Radius and an extremely low Amount. That is, with conventional USM the settings around 80% of the time will fall between 1.0 and 2.0 Radius and 250-500 Amount. With this hiraloam method, most of the time we'd be using Radii of 7-15 and Amounts of 70 or less. Doing something in between (medium Radius, medium Amount) doesn't seem to have anything going for it.

With High Pass, it's hard to select the right Radius, although the equivalent of Amount is easily achievable through the layer opacity setting. I prefer hiraloam USM because it's really easy to find the proper Radius and Threshold settings: we just set the Amount to 400% and adjust away. That makes it painfully evident what's being blurred out and what's being emphasized. After the Radius and Threshold are correct, then we drop the Amount to 60% or whatever seems indicated.

This kind of sharpening is usually *not* what we want to do. Conventional USM ordinarily gives better results, but there are exceptions:

*If there's nothing in the picture that has or requires an edge;
*If somebody else has sharpened the picture first;
*If the background is very busy and you're trying to avoid drawing attention to it;
*If you're really having a problem controlling noise with regular USM;
*If the objects you're mainly interested in are very large;
*If you're shooting for an "artistic" rather than highly focused look.

Examples of the difference between the two approaches.

1) A picture of a calm blue ocean. Hiraloam will work well here, creating soft ripples of lightness and darkness. Conventional USM works badly, creating needless edges.

1a) Same picture but with large waves breaking in the foreground. Without conventional USM we're unlikely to get the definition we need on the spray exploding away from the wave.

2) Close-in shot of a female face. As a rule we probably want to sharpen the black of CMYK, which will bring out the edges of the eyes and eyelashes. Or, we can sharpen overall with a high Threshold. Hiraloam usually won't work as well unless the face is quite large.

2a) Same shot but this time the model is heavily made up, especially eyeliner and mascara. In effect, the model has done the USM for us where it counts the most. We don't need to accentuate such a pronounced edge. Now hiraloam may make more sense to lighten the areas immediately around the eyes for a more dramatic effect. If we use this we probably have to use the sharpen tool (not the filter) on the center of the eyes because hiraloam isn't very effective there.

Nevertheless, images where hiraloam is outright better than conventional aren't all that common. Where it *is* more useful is in two combination cases:

*In certain classes of image, if you prepare two versions as best you can, one sharpened conventionally and one with hiraloam, a 50-50 or other split between the two sometimes works better than either one alone.

*More important, in CMYK there seem to be many images that can be improved by sharpening the black conventionally but the CMY using hiraloam. Low-key images are a good example but some faces work well also.

As pointed out by Stephen, this sharpening stuff can get very deep and eat up a lot of time. I currently recognize nine basic Photoshop sharpening techniques, of which conventional and hiraloam represent two. Plus, there are always combinations.

Dan Margulis


Date: Thu, 16 Jan 2003 05:08:41 -0800
From: "Stephen Marsh
Subject: Re: Advanced USM Options (was - Linear color space from AIM)

I wrote:

>> ii) Use a single duped layer and use the layer option blend if sliders to back off on the light halo (opt/alt click the white slider and split the left half over to the extreme left shadow point or lesser). Again, this is done on a luminosity fade sharpen layer before the blend if sliders are used. Pro - two layers, plus you can see the split intensity effect while you sharpen all in one go.<<

Small correction on this method, upon reflection this did not come out the way I intended.

The layer would be set to luminosity blend and not normal blend mode with a luminance fade. As the fade command is an extra processing step, time can be saved by using the luminance blend in the layer, as the blend-if sliders are performing the split halo intensity the luminance step may as well be in the layer itself. On smaller files this may not really matter, but if time counts then avoiding the fade command on a big file may be more of an issue. This also means that you can split the light half of the tonal range via sliders in the sharpen layer before sharpening, thus you can use higher USM settings than you probably would have previously, as the light halo preview is not overpowering the whole image. As the layer is being luminance blended, it is also safe to use the advanced blending to drop out unwanted channels (if applicable to image content).

When you alter halo intensity after sharpening, you may find that you could have used more USM than you previously thought - so if a split halo is performed after sharpening, it is probably best to oversharpen first and then to reduce the layer opacity if you find the setting too high once the light halo is backed off.

Stephen Marsh.


Date: Wed, 15 Jan 2003 20:11:22 -0200
From: "Claudio Portella"
Subject: Lecture material (was LAB Curves)

Dan,

And how about selling both the B/W handouts along with the lecture mp3 file? It would be very useful to folks who live far from U.S.

[ ]s
Claudio Portella -- www.claudioportella.com -- Rio de Janeiro, Brazil

----- Original Message -----
From: Dan Margulis

> I have handouts for all my sessions. However, 1) they're entirely B/W; 2)
> they are intended to supplement the lecture and would be difficult to
follow
> for somebody who didn't hear it; 3) at least in the case of the LAB session,
> the handout is somewhat of a stalking horse for some intended columns later on.


Date: Thu, 16 Jan 2003 10:53:23 EST
From: Dan Margulis
Subject: Lecture material (was LAB Curves)

Claudio writes,

>>And how about selling both the B/W handouts along with the lecture mp3 file? It would be very useful to folks who live far from U.S. >>

Even if the show sponsors agreed to allow this, which I doubt, the audio of the lecture wouldn't make much sense to the listener. The live demonstrations on screen are the key part.

In view of the interest I'll review the handout and if it seems to me it would be of any value at all to a reader who hasn't attended the lecture, I'll post it after the show in late February.

Dan Margulis


Date: Thu, 16 Jan 2003 10:31:37 -0600
From: Lori Sabo
Subject: Re: from an ex-pupil-sharpening, etc.

Dan:
> > There are imperfections in CCDs that can cause fringing or
> > weird colors.

This sounds valid to me. Unless you have something like the Foveon sensors that postition the R, G, and B sensing at the exact same location as each other there is a chance of some funky color fringing stuff especially with certain types of hi-frequency content. I actually saw a touch of this in one spot on a full-res (not interpolated) scan from our Betterlight Super-8K last night.

Of course, scanners can have the same problem, using the same types of ccd's.

Andrew:
> Very little if any when you work with quality equipment. The Betterlight
> scan back is a 3scanner on a stick2 but with no film, it1s producing vastly
> superior (and different files) that no scanner on this planet can compete
> with for pure quality of file. The files need to be handled differently
> because they are so different.

They are a whole nother world compared to a film scan! Truly cool. (And the added advantage that you have eliminated one extra color-perverting step by not shooting on film.) We just started using the Betterlight, we can come up with some watercolors digitized both ways if someone has a real interest in playing with them, since we also shot this last batch on 6x7cm.

Dan:
> > But the bottom line is, it's the character of the image, not the way it was
> > acquired.

But the acquisition method can affect the character of the final
captured image.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Lori Sabo


Date: Thu, 16 Jan 2003 09:53:10 -0700
From: Andrew Rodney
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Lori Sabo wrote:

> This sounds valid to me. Unless you have something like the Foveon
> sensors that postition the R, G, and B sensing at the exact same
> location as each other there is a chance of some funky color fringing
> stuff especially with certain types of hi-frequency content.

Only with one shot cameras that interpolate the color data. The Foveon (and older cameras like the Cat1s Eye which used 3 CCD1s and a beam splitter), didn1t have these issues. A 3/4 shot or true scanning camera is producing true color. And the fringing is getting to be less and less an issue with one shot. The difference between the Imacon 4040 in one shot and 4 shot is very difficult to see. So if the subject isn1t moving, there isn1t any reason not to use a multi-shot capture.

> Of course, scanners can have the same problem, using the same types of
> ccd's.

No, they produce true color as well (Trilinear CCD) which captures RG and B unlike single capture digital cameras with one chip which are just recording grayscale data and guessing about what the color should be. That1s where you get the weird fringing (Christmas tree effect).

> But the acquisition method can affect the character of the final
> captured image.

Yup, I agree!

Andrew Rodney


Date: Fri, 17 Jan 2003 01:41:29 +0100
From: Wim Melis
Subject: Sharpening low-res images

That was a lot of interesting reading. If I may sidetrack this a bit: there's one sharpening problem that's been bugging me for years. I've asked around a lot but never found a practical solution.

I stumbled across Corel Photopaint some time ago, and I discovered that its "Adaptive Sharpening" does a marvelous job at sharpening low-res images for web or cd-roms.

I'd love to get that same quality in Photoshop, but I've never been able to do so. Hence I'm still using Photopaint 8 for that final step.

The only answers I've ever received about this query involved lengthy Photoshop actions, a bunch of steps to perform without visual feedback along the way. That's too convoluted to be practical when doing large batches of images, I'm hoping for a quick method with immediate visual feedback instead.

Any specific advice on sharpening low-res images with PS?

Thanks, Wim


Date: Thu, 16 Jan 2003 21:36:12 -0500
From: Michael Demyan
Subject: RE: Sharpening low-res images

Hi Wim:

I have been using the Filter>Sharpen>Sharpen tool as my final step for
72dpi images posted on the web.

Mike Demyan


Date: Fri, 17 Jan 2003 03:37:02 -0000
From: Stephen Marsh
Subject: Re: Sharpening low-res images

Wim Melis wrote:

> I stumbled across Corel Photopaint some time ago, and I discovered that
> its "Adaptive Sharpening" does a marvelous job at sharpening low-res
> images for web or cd-roms.

> I'd love to get that same quality in Photoshop, but I've never been able
> to do so.

Wim, can you send me an unsharpened image and a adaptive sharpened image offlist, or sharpen one or two of the Photoshop 5 or 7 sample images so I can compare them against the original sample images that I have. Either offlist or via a download link or whatever.

Batch sharpening is going to be a trade off over quality vs. time, which is why adaptive methods proabably work for you.

In the past I have suggested high pass methods as these seek out high contrast with less variables than USM to worry about, but with nothing to compare against and the download from corel being 200mb for the entire suit it is hard to evaluate things. Even with more complex actions, batch accelerated processing of small files is not that slow.

Keep in mind that sharpening adds to jpeg file size, so you may want to explore edge masked USM or channel content masked USM and or smart blur or other selective bluring of low frequency areas while sharpening the high contrast areas as a trade off. This is why Adobe offer the blur feature in save for web, although it is not smart.

Stephen Marsh.


Date: Fri, 17 Jan 2003 12:35:25 -0000
From: Stephen Marsh
Subject: Re: LAB Curves

Andrew Rodney wrote:

> I1ve been asking Adobe to implement LCH
> corrections ala LinoColor and NewColor for years. LCH is VERY easy to learn
> and very powerful. It makes using LAB a cinch.

Wait a minute - sounds a bit like HSB mode, which Adobe axed after v2 or 2.5 (can't remember back that far<g>). Photoshop going full circle? Sure they are different things, but if this is a smaller gamut LAB without the opposing colour channel AB components and instead using chroma (saturation) and hue, it sounds very much like HSB or some such variation on the theme to me.

The old Mac Knoll HSB filter and the new Photoshop 7 HSB filters do not come close, as they do not offer previews and mode transforms loose a lot of data/visually change colours and tones which LAB often does not...but they can be good for some masks.

Due to the stated reasons for dropping HSB, I doubt that LCH would be adopted by Adobe - but one can hope.

Stephen Marsh.


Date: Fri, 17 Jan 2003 08:01:10 EST
From: Dan Margulis
Subject: Re: Re: LAB Curves

Stephen writes,

>>Wait a minute - sounds a bit like HSB mode, which Adobe axed after v2 or 2.5 (can't remember back that far<g>). Photoshop going full circle? Sure they are different things, but if this is a smaller gamut LAB without the opposing colour channel AB components and instead using chroma (saturation) and hue, it sounds very much like HSB or some such variation on the theme to me.>>

It is. LCH (Luminance, Chroma, Hue) is just HSB (Hue, Saturation, Brightness) with different names for the channels. What I was talking about was another space included in Linocolor, LAB (LH), which is a smaller-gamut version of LAB just as Ron was suggesting.

We all agree, however, that support for LCH or HSB or whatever you want to call it would be a significant improvement to add to Photoshop. Andrew Rodney is also right when he says that knowing LCH makes it much, much easier to learn LAB.

Dan Margulis


Date: Fri, 17 Jan 2003 08:11:45 EST
From: Dan Margulis
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Lori writes,

>But the acquisition method can affect the character of the final
>captured image.

Of course it can. However, the thread is about sharpening in Photoshop. Before you can do that you have to have a digital file. How it *came* to be a digital file is at that point of academic interest only.

Certain types of acquisition, for example, normally result in more noise than others. But no method is noise-free. Part of good sharpening is to avoid emphasizing noise. In doing that, we don't need to know what kind of camera shot it, we don't need to know what film was (or was not) used, we don't need to know who the photographer was, we don't need to know what year it was shot in. We need to know whether the file currently has noise, and if so, what kind.

Dan Margulis


Date: Fri, 17 Jan 2003 10:00:27 -0500 (EST)
From: rhansen
Subject: Re: Sharpening low-res images

Wim Melis wrote:

> I'd love to get that same quality in Photoshop, but I've never been able
> to do so. Hence I'm still using Photopaint 8 for that final step.

Without seeing specific examples from Photopaint, obviously, I don't know exactly what you're expecting. That being said, I generally start with web images at a larger size than they will be ultimately used. After correcting for color, I downsample to correct size and then apply Unsharp Mask with a radius of say, from .3 to .7 pixels (depending on image) and usually a threshold of 0 (the downsampling tends to eliminate a lot of the noise that would normally get sharpened with a 0 threshold).

This gives what I perceive to be pleasing results without halos or an oversharpened look. The images still compress efficiently too. One could set up an action with settings in this range to batch a bunch of photos, but as Stephen pointed out, there will be some tradeoff as one setting won't be optimum for all your images.

RJay Hansen


Date: Fri, 17 Jan 2003 08:47:28 -0700
From: Andrew Rodney
Subject: Re: Re: LAB Curves

Dan Margulis wrote:

> It is. LCH (Luminance, Chroma, Hue) is just HSB (Hue, Saturation, Brightness)
> with different names for the channels. What I was talking about was another
> space included in Linocolor, LAB (LH), which is a smaller-gamut version of
> LAB just as Ron was suggesting.

We need to be clear about LCH and size. There are actually TWO spec1s that LinoColor (which is the only product I know of that uses LCH) deals with. LinoColor had it1s own, home brewed type of LAB (which is hooked into LCH) they called LAB LH. Lab LH has a modified scale for the a and b axis to get more color shades in colors that really exist in scanned images while it ignores colors that don't exist in reality. As a consequence, if you open a Lab LH image in Photoshop as Lab, typically you get more saturated colors. So LAB LH and LAB (using LCH feedback) are different as are the sizes of the spaces.

Andrew Rodney


Date: Fri, 17 Jan 2003 08:52:37 -0700
From: Andrew Rodney
Subject: LAB, LAB LH and so on

Here is an interesting message I archived from the ColorSync list from Florian Süßl (who I think was at Heidlberg and worked with LinoColor). It1s some heavy duty color theory stuff but somewhat interesting in light of the discussion of LAB, LCH and 3smaller2 spaces for editing. Florina also dives into bit depth and the problems of editing in large colorspaces like Lab. While I1m sure there are those who would argue with Heidelberg that this may not be an issue, I1m not about to do that! These guys know their stuff. Shame they got out of the scanner business!

Andrew Rodney

Subject: Re: Photoshop versus LinoColor
Date: Thu, 23 Jul 1998 23:27:33 +0200
From: Florian Süßl
To: Peter.Bossens

Hi Peter,

Maybe I can add some information about your Lab discussion and Linotype Hell's (now Heidelberg PrePress or Linotype CPS) reason for choosing it as basic color model in LinoColor.

Peter, you mentioned that LinoHell might have some tricks under the hood...

The trick is pretty similar to the higher scanner data depth of 10 and more bit internally. The more bit the more shades between black and white the smoother transitions even after applying gradation corrections.

Compared to scanners and this problem Lab (8-bit Lab) is more like an 7bit image meaning that not 255 steps go from black to white but only 128! And you probably know about the problems of not having enough shades in a scan if corrections are needed.

Concerning Lab the Linocolor engineers really used and still use a trick to overcome this deficiancy of Lab compared to RGB: If you look at the Histogramm of an RGB-scan in Photoshop you will find that almost all color shades per channel (red, green, blue) goe from 0 to 255.

If you look at the same image in lab mode only the L channel covers almost all values defined by 8 bit meaning also 255 shades.

The problem of Lab are the a-(red-green chroma) and b-channel (yellow blue chroma): both only cover some 100 out of theoretically 255 shades: this phenomenon is based on the shape of the LAB-color model that is far off the ideal cube defined by L, a and b going from 0 to 255. compared to that cube the lab-model has a lot of "vacuum" around it's body meaning the theoretical values do *not* correspond with real colors.

An example: Along the gray axis from 0 (neutral black) to 255 (pure white) the a- and b- values are 0. At an L-value of 128 (mid lightness) a lot of a- and b-values exist, describing colors with that lightness. Close to 0 and 255 a- and b-values *cannot* have other values than 0. There simply is no color (red, green, etc.) close to these extreme Lightness values of pure black or pure white.

That's why scans described by Lab never cover the full range of theoretical values in the a- and b-channel. Or in other words: a- and b-values close to there extreme values -128 and 127 will never be used in a scan. The closer you come to either 0 or 255 in the L channel the less a- and b-values will show up.

Back to LinoColor: The Lino engineers decided to simply ignore these theoretical a- and b-values that will never be part of an image in order to use the 255 shades for the parts of the a- and b-axis that really exist. In Photoshop's color definition this is the range between -100 and +100 but not in 200 steps as in Photoshop: they cut it into 255 steps ending up with smoother gradations and lowering the risk of banding after corrections.

The name for that Lab definition is LAB LH (instead of LAB).

Florian Süßl


Date: Fri, 17 Jan 2003 09:04:08 -0700
From: Andrew Rodney
Subject: Re: Re: LAB Curves

Stephen Marsh wrote:
> Wait a minute - sounds a bit like HSB mode, which Adobe axed after v2
> or 2.5 (can't remember back that far<g>).

Not at all. LCH is so much easier to work with. The numbers are easy to understand (the scale for all three 3channels2) and LCH allows some really cool correction tools like those found in LinoColor or NewColor. I1ve been asking Adobe to put in a saturation curve. Works in LinoColor with LCH where you can pull a curve interface and increase or decrease saturation over the tonal scale of an image like you adjust tone using curves (edit only the C channel). Powerful. LinoColor also had a wonderful curve that would ONLY affect neutrals in the image. The selective color was powerful (but not really intuitive) because you could affect only the H channel. HSB in Photoshop is pretty much just an info palette.

Andrew Rodney


Date: Fri, 17 Jan 2003 11:36:17 EST
From: Dan Margulis
Subject: Re: LAB, LAB LH and so on

Andrew Rodney writes,

>>Here is an interesting message I archived from the ColorSync list from Florian Süssl (who I think was at Heidlberg and worked with LinoColor). It1s some heavy duty color theory stuff but somewhat interesting in light of the discussion of LAB, LCH and 3smaller2 spaces for editing. Florian also dives into bit depth and the problems of editing in large colorspaces like Lab.>>

This brings up an important tip for LAB users that I don't think has been posted here before. As Florian points out, much of the territory in the AB curves defines colors that are so brilliant that they don't exist in real life, or at any rate they can't be rendered accurately by any technology now known.

Everybody knows that LAB can easily create colors that are far out of the CMYK gamut, but some don't realize that it can also call for colors that are well beyond the capability of the monitor to display accurately. Photoshop gives no warning when this happens.

Oddly, this phenomenon is more dangerous to those whose destination is CMYK than RGB. If your destination is RGB, you can just convert the weirdo LAB file to RGB and at least now you'll have a file that matches what you saw on the screen when it was still in LAB. If your destination is CMYK, however, there's a major trap that needs to be avoided.

Ordinarily, converting an LAB file directly to CMYK will produce almost exactly the same file as converting it to RGB first and then to CMYK. When out-of-RGB-gamut colors are in play, though, the results can be wildly different, creating very unexpected CMYK results. For example, 100L43A48B is a fluorescent, electric orange-yellow that a monitor has zero chance of emulating. Instead, it shows what I would describe as a pale lemon yellow.

A copy of this LAB file, when converted to RGB, will look almost identical on screen. But when these two apparently identical files are converted to CMYK, they become very different indeed. On my system (your numbers may vary slightly depending on your Color Settings) the RGB version converts to 20M44Y, a reasonable approximation of the yellow that the monitor was showing. But when the LAB file is converted directly to CMYK, it goes to 34M38Y, which is not even a yellow at all, but a pink.

Solution: if you think you may have produced LAB colors too intense for the monitor to display, don«t convert it directly to CMYK. Convert it to RGB first. This will lock in colors that approximate what the monitor, however erroneously, is currently showing.

Dan Margulis


Date: Sat, 18 Jan 2003 03:40:39 -0800
From: Stephen Marsh
Subject: Re: LAB Curves

> Not at all. LCH is so much easier to work with. The numbers are easy to
> understand (the scale for all three 3channels2) and LCH allows some really
> cool correction tools like those found in LinoColor or NewColor.

Not having used Lino s/w I can't comment, but I have used HSB channels and edits in the old days in Photoshop when I first started, then the next thing I knew they were gone (with what seemed to me like a weak reason at the time, wish I could remember it now).

I do not find B hard to understand (easier than L), nor saturation. I would probably need a colour wheel poster to remember what each of the different exact hue angles were though.<g>

As mentioned, existing plugs for channel conversion are lacking in composite preview and suffer many visual composite colour and channel quantization errors.

I will take your word on the matter of LCH though, sounds nice.

> I1ve been
> asking Adobe to put in a saturation curve. Works in LinoColor with LCH where
> you can pull a curve interface and increase or decrease saturation overthe
> tonal scale of an image like you adjust tone using curves (edit only the C
> channel). Powerful. LinoColor also had a wonderful curve that would ONLY
> affect neutrals in the image.

The Creoscitex Eversmart supreme driven via oxygen software had all these features. You could pull a LS curves on the scan data and evaluate things either via RGB or CMYK while performing these LS curve corrections. It also has neutral S curve option checkbox if I remember correctly.

This is not to say that you can't write simulatd HS+Lum curves in Photoshop in RGB, but it does not seem to work as nice as a truly designed mode option - but there is always LAB if you really wan't to push things. Most of the time I use combined HS or color blend curves for this.

> The selective color was powerful (but not
> really intuitive) because you could affect only the H channel.

Intuitive from a colour science view, but perhaps not for a user used to RGB or CMYK only spaces where colour/tone are one.

The Oxygen s/w was great for a user of a prepress background, and unlike other scanner software that I know of you could even build alpha masks and do other things inside the software to perform selective edits, as well as the more regular options...all on full sized colour managed previews that could be softproofed etc.

I agree, there are many great concepts in hi end scanners which should migrate to Photoshop.

> HSB in Photoshop is pretty much just an info palette.

Now that is true, but some of us can still just remember the early days (although I only started with v2 so I am not that early a user). Photoshop does not have any real competition, but many other image editors offer HSB and HSV and other different but still useful modes.

Regards,

Stephen Marsh.


Date: Fri, 17 Jan 2003 11:42:32 -0500
From: Lippisch
Subject: Re: from an ex-pupil-sharpening, etc.

Dan,

With all due respect for your knowledge, I have to disagree with you on this subject.

As a commercial photographer and user of various digital SLR cameras and a BetterLight, I want to give support to Andrew's statements. I have worked with scanned 4x5, 2 1/4, and 35mm chromes, both from a high-end scanner and mid-range Nikon scanner. There is no question in my mind that the "character of the image" is greatly affected by "the way it was acquired".

In my experience digital camera files usually need much less sharpening with lower radius settings to look "finished" while film scans need much higher amounts with more effort to finesse them i.e.: fading USM to luminosity (which is a trick that I learned from your book Dan, thanks!).

Alex Lippisch


Date: Fri, 17 Jan 2003 12:11:20 -0500
From: "Jim Ray
Subject: RE: LAB, LAB LH and so on

Dan:

Just curious. You wrote

"For example, 100L43a48b is a fluorescent, electric orange-yellow that a monitor has zero chance of emulating. Instead, it shows what I would describe as a pale lemon yellow."

If a monitor has zero chance of reproducing it, and monitors typically have the largest output gamut and printing presses don't print in LAB anyway, how is it that we know that 100L43a48b is a fluorescent, electric orange-yellow? Is it because you know what the Gamut looks like, and where the formula should be located, or is there some output device out there I am overlooking?<puzzled>

By the way, this thread has been one of the most enlightening and enjoyable in some time. I support the notion of more education on LAB - either training seminars, magazine articels, or even another best-selling book (within the category of course - what chance does "LAB for Dummies" have against the Patricia Cornwall machine?)

Thanks for all your hard work and the knowledge you share.

Jim Ray a continuing student


Date: Fri, 17 Jan 2003 10:18:06 -0700
From: Andrew Rodney
Subject: Re: LAB, LAB LH and so on

Dan Margulis wrote:

> ... some don't realize that it can also call for colors that are
> well beyond the capability of the monitor to display accurately. Photoshop
> gives no warning when this happens.

> If your destination is RGB, you can just convert the weirdo LAB
> file to RGB and at least now you'll have a file that matches what you saw on
> the screen when it was still in LAB.

You lost me on those two paragraphs. In the first, say that there are colors that can't be displayed on the monitor and in the second you say the werdo LAB color that go to RGB will match what you saw on the screen when it was still in LAB. I1m confused.

> A copy of this LAB file, when converted to RGB, will look almost identical on
> screen.

That makes sense since the file has colors that can't be displayed. But for that matter, if you take a file outside display gamut (in say Adobe RGB or Wide Gamut RGB) and convert to a file that does fill into display gamut (sRGB), the previews look almost identical.

> Solution: if you think you may have produced LAB colors too intense for the
> monitor to display, don1t convert it directly to CMYK. Convert it to RGB
> first. This will lock in colors that approximate what the monitor, however
> erroneously, is currently showing.

Why convert? You can setup a soft proof. Open the LAB file and then pick the CMYK (or if you wish RGB) profile in Proof Setup and you can see what you1ll get if you convert from LAB to any space you have a profile for. You could set your display profile if you wish although I think it would be safer to pick an RGB Working Space.

Andrew Rodney


Date: Fri, 17 Jan 2003 10:30:32 -0700
From: Andrew Rodney
Subject: Re: LAB, LAB LH and so on

Jim wrote:

> If a monitor has zero chance of reproducing it, and monitors typically
> have the largest output gamut and printing presses don't print in LAB
> anyway, how is it that we know that 100L43a48b is a fluorescent,
> electric orange-yellow?

If you look at a gamut map of even the best displays next to that of many output devices you1ll see that the display isn1t necessarily larger and in many cases, there are colors in many, many output devices that fall outside display gamut. Display gamuts always show a very typical triangular gamut map. Output devices can have very weird gamut shapes that have areas that fall outside that triangle (even though much of the gamut is within). So the monitor isn1t the largest gamut output device you1ll encounter by a long shot. This is why Adobe placed the 3advanced Desaturate monitor colors by %2 in the color settings. It1s a real kludge! But the idea is to desaturate the display preview so you can see out of gamut colors change as you edit them. Of course that completely hoses the accuracy of the preview. But you do get to see out of gamut colors change...

> By the way, this thread has been one of the most enlightening and
> enjoyable in some time. I support the notion of more education on LAB -

With the death of Heidelberg's scanning division I think the LAB route is becoming a less used workflow. Dan is probably the only person who1s really pushing the LAB envelope that I know of. There are lots and lots of posts archived on the ColorSync list about the merits and pitfalls of LAB verses wider gamut RGB Working Spaces. Heidelberg and LinoColor had a pretty significant influence in Europe for years. But with that company out of the scanner business (and hence out of LinoColor/NewColor), I1m not sure what large company of hardware or software is really pushing for LAB. Again, if we could get Adobe to implement better LAB tools (using LCH specifically), this workflow could get a nice push. There are some tools in LinoColor using LAB and LCH that you just can1t do in Photoshop which is a shame.

Andrew Rodney


Date: Fri, 17 Jan 2003 13:27:44 -0500
From: Loring Palleske
Subject: Re: LAB, LAB LH and so on

Soft proofing can only show what the monitor is capable of producing. (where the gamuts overlap)

There are actually areas in CMYK that monitors cannot produce (I think in the pure CYAN area).

Regards,

Loring Palleske
Creative Imaging
1.877.279.2441
905.666.6647


Date: Fri, 17 Jan 2003 12:17:47 -0700
From: Andrew Rodney
Subject: Re: LAB, LAB LH and so on

Loring Palleske wrote:

> Soft proofing can only show what the monitor is capable of producing.
> (where the gamuts overlap)

Monitors can only show what monitors can show. So why convert a file to see what it will look like when you can do this with a soft proof faster and without possible data loss with quantization loss?

> There are actually areas in CMYK that monitors cannot produce (I think
> in the pure CYAN area).

Correct. Again, this is easy to see when you have a utility like Chromx's ColorThink that allows you to place one or more profiles over the other and view the overlap in 3D.

Andrew Rodney


Date: Fri, 17 Jan 2003 14:26:08 EST
From: Dan Margulis
Subject: Re: LAB, LAB LH and so on

Andrew Rodney writes,

>>You lost me on those two paragraphs. In the first, say that there are colors that can't be displayed on the monitor and in the second you say the weirdo LAB color that go to RGB will match what you saw on the screen when it was still in LAB. I'm confused.>>

If you're calling in LAB for a color that the monitor can't display, it still has to display *something*. That something is what you are going to get if you copy the file and convert it into RGB. At that point, you'll have two files that look almost exactly alike but are in fact as much alike as lobster bisque and motor oil.

IOW, suppose you have an LAB file that calls for something ridiculous like 90L100a100b, brilliant orange. Make a copy and convert it into, say, Colormatch RGB. Photoshop does the best it can to create this impossible color, 255r60g0b. It *looks* almost exactly like the LAB file did, because the monitor can't display the color that LAB is calling for and is improvising.

So, now you have two seemingly identical oranges. But if you convert them both into CMYK, you're likely to get a really big surprise, because the Colormatch RGB file is really calling for 64L65a70b, which is a far cry from 90L100a100b. On my system the RGB file converts to 72m97y but the seemingly identical LAB file goes to 65m81y which is nothing like it. The RGB file gives me much more of what I am expecting.

>>Why convert? You can setup a soft proof. Open the LAB file and then pick the CMYK (or if you wish RGB) profile in Proof Setup and you can see what you1ll get if you convert from LAB to any space you have a profile for. >>

That's right, but I won't know whether it's the best I can do unless I make a copy of the file, convert it to RGB, and look at it with the same proofing setup. Every time I've done this so far, the RGB version is considerably better, as it is in the above example.

Dan Margulis


Date: Fri, 17 Jan 2003 14:29:55 EST
From: Dan Margulis
Subject: Re: LAB, LAB LH and so on

Jim writes,

>If a monitor has zero chance of reproducing it, and monitors typically
>have the largest output gamut and printing presses don't print in LAB
>anyway, how is it that we know that 100L43a48Bb is a fluorescent,
>electric orange-yellow?

Because 100L means absolutely the brightest possible--if you shoot directly into the sun, you can't get brighter than 100L. And yet there is a very strong color component. It favors magenta in the A and yellow in the B, in roughly the same proportions as a shot directly into the sun would, but far stronger.

So, we're asking for a color that's as bright as the sun and the same hue as the sun, but far more colorful than the sun is. "Fluorescent, electric orange-yellow" is the best description I can come up with for that color, and yes, I am quite sure the monitor has no chance whatever of portraying it accurately.

BTW, monitors *don't* have the largest gamut in colors like this. There do exist fluorescent inks that can come reasonably close to hitting this color, more so than a monitor can. If this were more of a yellow and less of an orange, even straight CMYK would be able to do better than the monitor.

Dan Margulis


Date: Fri, 17 Jan 2003 12:05:01 -0700
From: Chris Murphy
Subject: Re: LAB, LAB LH and so on

Dan Margulis wrote:

> Oddly, this phenomenon is more dangerous to those whose destination is CMYK
> than RGB. If your destination is RGB, you can just convert the weirdo LAB
> file to RGB and at least now you'll have a file that matches what you saw on
> the screen when it was still in LAB. If your destination is CMYK, however,
> there's a major trap that needs to be avoided.

If it's either RGB or CMYK output, the trap is pretty major. Use of Proof Preview is highly underrated, and would largely solve the gotcha.

> Ordinarily, converting an LAB file directly to CMYK will produce almost
> exactly the same file as converting it to RGB first and then to CMYK. When
> out-of-RGB-gamut colors are in play, though, the results can be wildly
> different, creating very unexpected CMYK results. For example, 100L43a48b is
> a fluorescent, electric orange-yellow that a monitor has zero chance of
> emulating. Instead, it shows what I would describe as a pale lemon yellow.

> A copy of this LAB file, when converted to RGB, will look almost identical on
> screen. But when these two apparently identical files are converted to CMYK,
> they become very different indeed. On my system (your numbers may vary
> slightly depending on your Color Settings) the RGB version converts to
> 20m44y, a reasonable approximation of the yellow that the monitor was
> showing. But when the LAB file is converted directly to CMYK, it goes
> to 34m38y, which is not even a yellow at all, but a pink.

Note first that I support the conclusion.

Second, I question if 100L*, 43a*, 48b* is even a real color. Just like I question 0L*, 90a*, -128b* is a real color. That Photoshop lets you define such colors, they shouldn't exist because at L*=0 the color is black, yet those are the values you get for the blue primary in ProPhoto RGB. Not only should it be black of L*=0, but it's actually an encodable value that puts it outside of the CIE xyY chromaticity diagram, so it's not even a color, just an encoded value. L*=100 should be white and extreme a* and b* values just don't exist, yet Photoshop does allow their input (it's the difference between LAB as a real color space and LAB as an encoded color space).

I think what's happening in the case of going directly to an output profile (CMYK or RGB) is that these profiles based on tables are equipped for dealing with out of gamut colors. They deal reasonably well (considering) when out of gamut colors fall within the range assumed by the profile building application. This doesn't occur with a matrix based RGB profile (ColorMatch RGB, Adobe RGB, etc.) because they weren't built with assumptions about the source space. So you get true clipping, not gamut compression.

When I use an RGB output device profile, I get the same results (visually) as if I had converted it to CMYK. So this isn't so much an RGB vs. CMYK issue as it is an editing space profile vs. output device profile issue.

> Solution: if you think you may have produced LAB colors too intense for the
> monitor to display, donÕt convert it directly to CMYK. Convert it to RGB
> first. This will lock in colors that approximate what the monitor, however
> erroneously, is currently showing.

Yes. But I've also noted that the direct CMYK conversion converts the LAB colors more closely. So it's a matter of whether you want what you see on your monitor, or if you want what's actually in the file. To decide, use Proof Preview which will show you what you're going to get if you do a direct conversion to CMYK.

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


Date: Fri, 17 Jan 2003 14:49:25 -0500
From: Loring Palleske
Subject: Re: LAB, LAB LH and so on

At 100L is not everything white?

Regards,

Loring Palleske
Creative Imaging
1.877.279.2441
905.666.6647


Date: Fri, 17 Jan 2003 13:00:53 -0700
From: Andrew Rodney
Subject: Re: LAB, LAB LH and so on

Dan Margulis wrote:

> If you're calling in LAB for a color that the monitor can't display, it still
> has to display *something*.

Well sure. Photoshop has to always take anything that isn1t RGB (in this case LAB) and do a conversion to the screen for RGB. Same with CMYK. Photoshop only displays in RGB.

> That something is what you are going to get if
> you copy the file and convert it into RGB.

I1m not sure about that. When you convert to RGB, you can specify an awful lot of RGB spaces. Adobe RGB or sRGB? The numbers from LAB to those two spaces are going to produce different resulting RGB values no? Both previews from LAB may look the same but the numbers should be different. With PS71s Proof Color option in the Info Palette, you can setup any RGB profile you want and get the resulting values before you convert (but CMYK isn1t in the equation yet).

> At that point, you'll have two
> files that look almost exactly alike but are in fact as much alike as lobster
> bisque and motor oil.

I agree.

> IOW, suppose you have an LAB file that calls for something ridiculous like
> 90L100a100b, brilliant orange. Make a copy and convert it into, say,
> Colormatch RGB. Photoshop does the best it can to create this impossible
> color, 255r60g0b. It *looks* almost exactly like the LAB file did, because
> the monitor can't display the color that LAB is calling for and is
> improvising.

Again, that makes sense. The only thing I1d add is that if I have ColorMatch RGB and Adobe RGB setup as soft proofs and have the info palette set to reflect this, I can toggle the proofs and see the resulting numbers before doing the conversion. I realize the previews will look the same. The numbers will be different.

> So, now you have two seemingly identical oranges. But if you convert them
> both into CMYK, you're likely to get a really big surprise, because the
> Colormatch RGB file is really calling for 64L65a70b, which is a far cry from
> 90L100a100b. On my system the RGB file converts to 72m97y but the seemingly
> identical LAB file goes to 65m81y which is nothing like it. The RGB file
> gives me much more of what I am expecting.

OK, that1s clear and I1d expect that because this orange in ColorMatch is a vastly different number than what the LAB file converted to another RGB space would be from LAB. This may illustrate the profound effect the resulting RGB Working Space you pick from LAB plays when going to CMYK (or for that matter, any resulting output space).

> That's right, but I won't know whether it's the best I can do unless I make a
> copy of the file, convert it to RGB, and look at it with the same proofing
> setup. Every time I've done this so far, the RGB version is considerably
> better, as it is in the above example.

It sounds like what you want is a way to work in LAB and see what the values would be in a certain RGB Colorspace and THEN what the CMYK numbers would be from that space. A three way conversion number feedback. With the current feature set, you can open a LAB file and get LAB to CMYK or LAB to RGB values from the Proof setup but not LAB to RGB to CMYK values. But what you can do is open a LAB file and set one of the info palettes to Proof Colors and load your CMYK output profile in the proof setup AND in the 2nd info palette ask for RGB color (it will give you the RGB Working Space set in your color preferences). If you go into color settings and change the RGB Working Space, that 2nd palette will update the values from LAB to that RGB space if you converted. Now the problem. The CMYK numbers don1t change when you toggle from undo to redo. So it appears the CMYK numbers are LAB to CMYK, not LAB to RGB to CMYK. That latter behavior might be useful to have as long as it1s easy to specify the behavior for the feedback.

The 3problem2 is that we have files with vastly different sets of numbers that can all preview the same. I1m not sure that1s necessarily wrong but I think what you want is to get the values (or resulting values) for files that look the same. Converting certainly accomplishes this but it might be nice if you could get the values you want while the file is in LAB (or any colorspace) without actually converting. There doesn1t seem to be a reason Photoshop couldn't do this.

Andrew Rodney


Date: Fri, 17 Jan 2003 14:53:17 EST
From: Dan Margulis
Subject: Re: LAB, LAB LH and so on

Chris Murphy writes,

>Second, I question if 100L*, 43a*, 48b* is even a real color. Just like
>I question 0L*, 90a*, -128b* is a real color.

The first one probably isn't but might be, the second one certainly is fantasyland.

>Yes. But I've also noted that the direct CMYK conversion converts the
>LAB colors more closely. So it's a matter of whether you want what you
>see on your monitor, or if you want what's actually in the file.

This is like asking which artist has drawn the most accurate picture of a Martian. If the color is imaginary it's real hard to decide which CMYK equivalent is closest to it. I think that almost invariably one would prefer the imaginary color first converted to RGB and then to CMYK. Once in RGB, at least it's a real color that the monitor can tell us something about. If we convert an imaginary color directly into CMYK God knows what we may get, but I don't think either of us does.

Dan Margulis


Date: Fri, 17 Jan 2003 16:52:34 -0600
From: Ron Bean
Subject: Re: LAB, LAB LH and so on

Andrew Rodney writes:

>From: Florian Süssl

>Along the gray axis from 0 (neutral black) to 255 (pure white) the a- and
>b- values are 0.
>At an L-value of 128 (mid lightness) a lot of a- and b-values exist,
>describing colors with that lightness.
>Close to 0 and 255 a- and b-values *cannot* have other values than 0. There
>simply is no color (red, green, etc.) close to these extreme Lightness
>values of pure black or pure white.

OK, that's another way of saying that pure white and pure black have zero saturation by definition, and max saturation can only occur at the midpoint of the L curve.

I knew this but I hadn't thought about the implications for LAB. So if you're messing around with the A and B curves near the zero point, you could easily drive the highlights and shadows out of any real gamut, and as Dan points out, you don't know where they'll end up when you convert back to a printable colorspace.

I suppose you could use a mask to limit the range of L values being affected by the correction (or use some other colorspace for that kind of correction).

Am I on the right track here?


Date: Sat, 18 Jan 2003 08:20:18 EST
From: Dan Margulis
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Alex writes:

>With all due respect for your knowledge, I have to disagree with you on
>this subject.

On the contrary, nothing you say in your post disagrees with anything I said. Neither, despite all the bile, does anything Andrew Rodney's post contains.

>There is no question in my mind that the "character of the image"
>is greatly affected by "the way it was acquired".

There is no question in anybody's mind about this. However, once the "character of the image" is determined, the way it was acquired is no longer relevant.

You apparently believe that digital captures require less sharpening than scans. I don't think I agree (I think it would be more accurate to say that they don't require as high a Threshold, which may reduce the Amount and Radius needed), but I won't argue the point.

But even if we limit it to Threshold, the information is useless unless we can say absolutely, unconditionally, without exception that every single digital capture requires a lower Threshold than every single scan. In any other scenario, we are compelled to open the image and examine it to see what kind of Threshold is required. And the exact same factors will apply in this decision regardless of the method of capture.

Image editing is like text editing. If we are assigned to edit drafts of two articles and we are told one is written by someone whose first language isn't English, we probably have some preconceptions about how the two will differ. We may expect that that one needs more editing than the other, or that it may contain certain types of error that aren't characteristic of native speakers. But there's no guarantee that either of these preconceptions is correct. Furthermore, once we have the two drafts in hand, we treat them exactly alike. If a word is misspelled, or a phrase makes no sense, we don't ask which writer wrote it. We just fix it and move on.

Similarly: if you provide me with a file you shot on a very expensive digital camera, and somebody else gives me a scan of something shot by an amateur on a disposable camera, I have certain preconceptions about how the two may differ. But once the two are open on my screen, those preconceptions are history. If the images share the same characteristics they'll be sharpened in the same way. If they don't, they'll be analyzed in the same way and each one will be sharpened differently. But where each came from is, at the point, irrelevant. They're just a couple of images that need sharpening.

Dan Margulis


Date: Sat, 18 Jan 2003 08:20:50 EST
From: Dan Margulis
Subject: Re: LAB, LAB LH and so on

Andrew Rodney writes,

>1m not sure about that. When you convert to RGB, you can specify an awful
>lot of RGB spaces. Adobe RGB or sRGB? The numbers from LAB to those two
>spaces are going to produce different resulting RGB values no? Both previews
>from LAB may look the same but the numbers should be different.

Yes.

>It sounds like what you want is a way to work in LAB and see what the values
>would be in a certain RGB Colorspace and THEN what the CMYK numbers would
>be from that space. A three way conversion number feedback. With the current
>feature set, you can open a LAB file and get LAB to CMYK or LAB to RGB
>values from the Proof setup but not LAB to RGB to CMYK values.

Correct. That's what we want.

Dan Margulis


Date: Sat, 18 Jan 2003 08:18:51 EST
From: Dan Margulis
Subject: Re: LAB, LAB LH and so on

Loring writes,

>At 100L is not everything white?

And Ron adds, quoting Florian Süssl,

>Close to 0 and 255 a- and b-values *cannot* have other values than 0. There
>simply is no color (red, green, etc.) close to these extreme Lightness
>values of pure black or pure white.

These statements are only true in our RGB or CMYK-centric world. If 100L0a0b is defined as the brightest possible white the only sensible way to portray it is as 255r255g255b or 0c0m0y0k. But RGB and CMYK are messed up, in that pure white is absolutely the brightest combination that can be made. Try to add any hue to 255r255g255b, or any ink to 0c0m0y0k, and you may get color but the result has to be darker than the pure white.

Therefore, if you can imagine something that's as bright as pure white yet carries a color, it is *by definition* out of both the RGB and CMYK gamut. And yet such things exist. I don't know of any pure whites that I would consider to be brighter than the sun, or than a blue neon light, or than a red laser. And yet all these have distinct colors. So, contrary to what Florian says, in nature there can be, and are, things that are 100L and not 0a0b. But, in agreement with what he says, there isn't any possibility of reproducing them correctly. LAB can define them but RGB and CMYK can't.

How many eons of photographer time has been spent trying to perfect pictures of sunsets? The reason is suggested by Loring's question. And the answer is, CMYK and RGB require 100L to be accompanied by 0a0b. The catch is, nobody clued God in that He was supposed to play by any such rules.

Dan Margulis


Date: Sat, 18 Jan 2003 04:59:50 -0800
From: "Stephen Marsh"
Subject: Re: LAB Curves

I wrote:

>> I agree, there are many great concepts in hi end scanners which should migrate to Photoshop.

Hah - I keep forgetting that Photoshop was originally a bit of scanner software.<g>

Stephen Marsh.


Date: Sat, 18 Jan 2003 12:24:48 -0700
From: Ron Kelly
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Dan Margulis wrote:

But where each came from is, at the point,
> irrelevant. They're just a couple of images that need sharpening.

Since this thread is all about sharpening I thought I'd inject ask this question which I've always been curious about.

If I shoot some Velvia (fine grain reversal film) with a high quality optic, like, say, a Zeiss, and I'm very careful and expose it just so and use tripod, etc., I'm rewarded with a very sharp piece of film.

If I take that same film and photographically enlarge it, again using first rate equipment and careful technique, I can get a very sharp print. All the enlarging was done optically; there was no "sharpening" involved.

Save for a small variation due to potentially a contrast increase, it can be said that no sharpness is added by photographic printing. If anything, some sharpness will more likely be lost due to enlargement.

However, if I scan that same piece of film using a first rate scanner, or if I used a digital back of the best quality for the Hasselblad for the image capture in the first place, I get a file that is still soft.

Or is it? Is there any type of scanner/digital capture device that reduces or eliminates the need to sharpen the result?

Let's assume for sake of argument that we're talking about using the full resolution of the device, and not downsampling, ie if the device captures 3000 pixels wide at it's best/full quality, you view or print it that way.

Ron Kelly


Date: Sat, 18 Jan 2003 13:13:27 -0700
From: Andrew Rodney
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Ron Kelly wrote:

> However, if I scan that same piece of film using a first rate scanner,
> or if I used a digital back of the best quality for the Hasselblad for
> the image capture in the first place, I get a file that is still soft.

You1ve taken a continuous tone original (the film) and converted it to a mosaic of square pixels. Any edge that was anything but a straight line is now less sharp since all you1ve done is make a grid. So some sharpening is necessary because the process of digitizing inherently softens the appearance of the file. This is one reason why USM is subject as well as resolution dependant. You1d handle a scan of a building differently than a scan of a portrait even if both files were the same size!

> Let's assume for sake of argument that we're talking about using the
> full resolution of the device, and not downsampling, ie if the device
> captures 3000 pixels wide at it's best/full quality, you view or print
> it that way.

This is why sharpening is resolution dependant. When you scan your 4x5 at 300ppi verses 5000 ppi, and view both at 100% in Photoshop, you1ll see a pretty significant difference in the mosaic rendering all those round-ish edges! You can1t possibly sharpen the two files the same.

Photograph some finely ruled lines like the U.S. Resolution chart and scan it at 300ppi and 5000ppi (or shoot it with a digital camera and downsize one version). You1ll see a big difference in how MANY pixels are used to produce a round object in the target. The higher rez image will have far more pixels defining a round shape using what is essentially a series of straight lines. Pixels are solid and square. If we had pixels that could have the top half white and the bottom half black, think how we could make a smooth curve. But we can1t do this. The digital file is a series of squares trying to represent curves. Due to this, we lose some sharpening when we convert our continuous tone world into a world of squares.

To some degree, grain could be said to be the same although in a different way (each grain in film may be a certain shape but you don1t see a mosaic even with the strongest lope).

In the end, we usually have to take this data and end up with dots to print the darn thing. So USM is output dependant. The dots you1d get from a halftone device are vastly different than the dots off a LightJet or an Epson printer. So USM/sharpening is resolution, subject and output dependant.

Andrew Rodney


Date: Sat, 18 Jan 2003 21:38:29 +0100
From: Henk Gianotten
Subject: Lab and LCH are extremely important to cover

Dan Margulis wrote on 14 Jan 2003 02:21:35:

>There are four full chapters on LAB or LAB-like maneuvers in Professional
>Photoshop and that's about all the room I think I can justify. There are so
>many things that LAB does well that I have toyed with the idea of writing a
>book about LAB only, with the idea of making it more accessible to the
>masses. But, I just don't think there's a big enough market to make it
>worthwhile.
---------------------------------------------------------------------------------------------------------
Hi Dan,
I agree that many professionals in graphic arts and photography (capturing and editing) don't have much interest in Lab.

However, as you learned recently, Lab transformations are very sophisticated and far more effective than other CMYK- and/or RGB-related tools.

By proving the advantages, showing real-life results and explaining what it does the interest will increase.

Such a book will improve the quality of skilled retouchers, prepressers, photographers and operators who have to handle pictures in databases. They sometimes let the system perform extensive color-, sharpness and tonal corrections without knowledge about the influence.

I was involved in several test to compress images in an Lab-workflow. We had excellent results but were unable to explain the advantages to the majority of the existing users.

For them Lab was just another magic trick. They disliked it due to the lack of knowledge and (non-scientific) explanations.

And LCH (LAB-like maneuvers?) was another set of magic.The best system to correct generic images.

Some German researchers (like Dr Sheran Tatari) made clever presentations but failed to convince the audience. Just a few skilled scanner operators and image-editors understood the impact of such a workflow and system. Most marketing guys were not interested in long-term (and costly) activities to promote it.

It looks, like a lot of developments, that such a change in attitude of the users takes far more time than we expect (or hope).

I would strongly suggest to write such a book, to promote Lab and explain the advantages of LCH.

The market will develop; I herewith order the first copy of your book covering that..

Regards, Henk


Date: Sat, 18 Jan 2003 16:08:52 -0500
From: Loring Palleske
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Ron Kelly wrote:

> Save for a small variation due to potentially a contrast increase, it
> can be said that no sharpness is added by photographic printing. If
> anything, some sharpness will more likely be lost due to enlargement.

But sharpness could be added using an unsharp mask - it was originally a darkroom process. Most of the stuff photoshop can do started with a darkroom / retouching / manual process.

The degree of control is just much better and who could live without the undo command.

Regards,

Loring Palleske
Creative Imaging
1.877.279.2441
905.666.6647


Date: Sat, 18 Jan 2003 16:10:38 -0600
From: Ron Bean
Subject: Re: Re: from an ex-pupil-sharpening, etc.

I think you'll find that the amount of sharpening depends on the resolution of the output device, not the input device. A 100 lpi halftone will require more sharpening than a 200 lpi halftone, and scanning at a higher resolution won't help because the extra resolution is thrown away at the output stage. Even with a non-halftone printer, you're still printing in rows and columns.

I think it's possible that a *very* high resolution output device might need little or no sharpening (assuming the input file has enough resolution also). But I don't know if it would be cost effective. I once checked a bunch of expensive art books with a halftone screen checker, and none of them were above 150lpi, even though 200lpi printing is available.

If your output is to a film recorder, the need for sharpening will depend on how much you plan to enlarge it when you print from the film.

I don't think that eliminating sharpening is necessarily a worthy goal. Sharpening is only bad if it's done badly :-)


Date: Sat, 18 Jan 2003 17:16:40 EST
From: Dan Margulis
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Ron Kelly writes,

>Is there any type of scanner/digital capture device that
>reduces or eliminates the need to sharpen the result?

No, and if there were, you would still need to sharpen if you intended to print it by any nonphotographic process.

Transitions get hosed by any digital capture process and then get hosed again by any kind of output. Where two unlike colors butt, in your film you'll have a line of transition so fine that no capture device can grab it. Instead, it will create pixels that belong to neither color, making for a blurry effect. It gets worse during printing, since the halftone dots are even larger in comparison to the transition in the film.

For realism, the transitions have to be re-emphasized. As the thread has indicated, there are several distinct strategies that can accomplish this. But you do have to do *something.*

Dan Margulis


Date: Sat, 18 Jan 2003 17:06:18 -0700
From: Ron Kelly
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Andrew Rodney wrote:

> You've taken a continuous tone original (the film) and converted it to a
> mosaic of square pixels. Any edge that was anything but a straight line
> is now less sharp since all you?ve done is make a grid. So some
> sharpening is necessary because the process of digitizing inherently
> softens the appearance of the file.

Thank-you Andrew. That answer is so simple I don't know why I didn't think of that myself.

Ron Kelly


Date: Sat, 18 Jan 2003 17:02:29 -0700
From: Ron Kelly
Subject: Re: Re: from an ex-pupil-sharpening, etc.

Ron Bean wrote:

> I don't think that eliminating sharpening is necessarily a worthy
> goal. Sharpening is only bad if it's done badly :-).

Right you are. But my goal is not to eliminate sharpening, just understand how/why it's neccessary. This understanding will ultimately make my work better (I hope).

Ron Kelly


Date: Sun, 19 Jan 2003 01:43:36 -0000
From: Stephen Marsh
Subject: Image Sharpness

Norman Koren has a fantastic site, a couple of pages which came to mind on the topic of loosing sharpness from the original can be found here:

http://www.normankoren.com/Tutorials/MTF.html

http://www.normankoren.com/Tutorials/MTF5.html

Just explore the whole site, one can spend a lot of time here.

Stephen Marsh.


Date: Sun, 19 Jan 2003 15:24:32 -0000
From: "Herbert Gibson
Subject: Color Shift on conversion to Lab

Amateur photographer, just recently signed up to the group.

I'm interested in the possibilities of working in Lab. I've calibrated my monitor (Spyder) and use profiled files but have found that there is a strong color shift to Blue when I change to Lab (or to CMYK). What's going on please?


Date: Sun, 19 Jan 2003 16:08:19 -0000
From: "danzimmerman
Subject: Interesting Sharpness test

Interesting Sharpness Test

There is an interesting sharpness test over at DPR retouch forum. Results were not what was expected. The name of the thread is Sharpening Methods Compared.

http://forums.dpreview.com/forums/read.asp?forum=1006&message=4208293

Dan Zimmerman


Date: Sun, 19 Jan 2003 12:40:05 -0500
From: Jim Rich
Subject: Re: Color Shift on conversion to Lab

Herbert Gibson wrote:

> I'm interested in the possibilities of working in Lab. I've
> calibrated my monitor (Spyder) and use profiled files but have found
> that there is a strong color shift to Blue when I change to Lab (or
> to CMYK). What's going on please?

Herb,

While your message says nothing about your past experience with RGB or CMYK color spaces for editing an image, your post has stirred my curiosity as to why you want to work in LAB. I find that you can easily color correct and get great results from images in RGB and CMYK without too much head banging. You can also color correct images in Lab but, as many will admit, LAB is not very intuitive. There are valid reasons to work in LAB but it you want to remove a blue color cast why not do it in RGB or CMYK. This is by far easier. Or is there something else going on here I have not understood?

As for what is going on when you convert to LAB or to CMYK, well...consider these questions.

What is the source color space that the image is starting in? The scanners RGB? Unknown RGB or CMYK? A working space such as Adobe RGB? ???

Where did the images come from that have profiles attached to them?

How are you making the change from your source color space to LAB or CMYK?

Are you using the Photoshop Mode change? Or are you using Convert to Profile? If you are using the MODE, are you using the default in Color Settings?

If you are using the default Color Settings then you are most likely using the wrong printer profile. To get the best results you would need a custom printer profile.

And, where is the blue color shift, on the monitor or on the final print?

Actually, as I have re-read your post and my response, I sounds like you might be using the monitor profile you created with your Spyder as a Working Space. Where have you placed your monitor profile? Is it in the correct folder or directory so your monitor can call on it properly.

Jim Rich


Date: Sun, 19 Jan 2003 12:27:47 -0800
From: Jan Steinman
Subject: Re: from an ex-pupil-sharpening, etc.

From: Ron Kelly

>if I scan that same piece of film using a first rate scanner,
>or if I used a digital back of the best quality for the Hasselblad for
>the image capture in the first place, I get a file that is still soft.

This is a basic aspect of quantization. Whenever you "sample" information, you are necessarily excluding some of it. The process of sampling limits high-frequency information to 1/2 the sampling aperture -- the Nyquist limit.

Few scanners are capable of sampling at the Nyquist limit of Velvia. For one thing, you just can't squirt enough energy through such a tiny aperture. The very best drum scanners have 3 micron apertures, which require 1,000 watt lamps, and you can only pull about 1,700 watts from the wall.

In addition, CCD scanners have a whole bunch of other constraints thrown in, such as dark-current noise (which "softens" shadow detail) and adjacent pixel leakage (which "softens" local contrast via "blooming").

Having worked as signal processing engineer years ago, I'm amazed this stuff works as well as it does! And it will continue to get better. But for the present, a combination of theoretical and engineering limitations do cause sampled images to be "softer" than photochemical images.

--
: Jan Steinman -- nature Transography(TM): http://www.Bytesmiths.com>
: Bytesmiths -- artists' services: http://www.Bytesmiths.com/Services


Date: Mon, 20 Jan 2003 10:33:16 -0000
From: Stephen Marsh
Subject: Re: Color Shift on conversion to Lab

> I'm interested in the possibilities of working in Lab.

There are many pros/cons with LAB.

Searching the ColorSync archive as Andrew Rodney recently suggested will bring forth the mainstream view (both facts and opinions). This is a huge wealth of knowledge just sitting there, rather deep stuff (after reading the first 500 search results on LAB I got a bit tired and my attention wandered, there are still many search hits left to read<g>).

http://lists.apple.com/ (look for colorsync users archive)

Dan has written a few things on LAB in the past which may help:

https://www.ledet.com/margulis/ACT_postings/ACT-LABLoss.html

https://www.ledet.com/margulis/LABCorrection.pdf

http://ep.pennnet.com/Articles/Article_Display.cfm?Section=Archives&Subsection=Display&ARTICLE_ID

http://ep.pennnet.com/

https://www.ledet.com/margulis/ACT_postings/ACT-blurring.html

> I've calibrated my monitor (Spyder) and use profiled files but have found
> that there is a strong color shift to Blue when I change to Lab (or
> to CMYK). What's going on please?

It can depend on the source and the space you are converting to...

Blue is a known problem colour for CMYK.

There are blues and other colours which can shift when converted into ICC LAB (Photoshop LAB mode) as it is not possible to represent these colours in LAB. I once posted an example of this to the list of a blue sky which changed hue.

It would be REALLY good if you can point us to a crop of the blue that suffers in the change to LAB.

Many LAB/HSB type edits can be simulated directly in RGB/CMYK in Photoshop by using an adjustment layer or a regular layer in luminosity blend mode or colour blend mode (or hue or saturation). There is also the option to use the fade command directly after an edit (if there is no layer mask on the layer).

As Dan recently mentioned, LAB works very well for the shadows when sharpening (you could make a gamma 3 RGB workspace and perhaps have less loss in the transform and bypass ICC LAB mode).

The option also exists to dupe the RGB file into LAB, sharpen - back to RGB and isolate the shadows only and then mask these into the rest of the image. This way your colours will not suffer and you target the sharpening fromt the LAB space to where it is most effective. The lower layer would be sharpened in regular 1.8 or 2.2 gamma RGB WS.

Stephen Marsh.


Date: Sun, 19 Jan 2003 19:17:51 -0000
From: Herbert Gibson
Subject: Re: Color Shift on conversion to Lab

Jim Rich wrote:

> ...why you want to work in LAB. I find that you can easily color correct and get great results from images in RGB ...

I was interested more in L channel possibilities, sharpening, etc. Would still like accurate color for this.

> You can also color correct images in Lab but, as many will admit, LAB is not very intuitive...

Too scary for me at this stage!

> What is the source color space that the image is starting in? Where did the images come from that have profiles attached to them?

I have 3 test images: One from Andrew Rodney is ColorMatch. This converts without any visible change. A Fuji test image is sRGB - it shows a definite blue shift. Finally, one from Pixl in AdobeRGB converts with a very strong blue cast.

> How are you making the change from your source color space to LAB or CMYK?

Image / Mode / Lab (Photoshop 6)

> If you are using the MODE, are you using the default in Color Settings?

I've actually got Bruce RGB as the RGB workspace but I instruct Photoshop to use the embedded profile as each file opens. I assumed that Lab is an absolute reference not linked to the settings.

> If you are using the default Color Settings then you are most likely
> using the wrong printer profile.
> And, where is the blue color shift, on the monitor or on the final print?

I've not been printing - the effect is on the monitor.

> Where have you placed your monitor profile? Is it in the correct
> folder or directory so your monitor can call on it properly.

I think it's OK. It's correctly specified in the graphics card SW (Matrox 450) and visibly loads at start-up.


Date: Mon, 20 Jan 2003 10:16:30 -0500
From: Jim Rich
Subject: Re: Re: Color Shift on conversion to Lab

Herb,

Thanks for filling in some of the blanks.

I think to figure out what is going on with your system you have to get it all under control so you can learn each components behavior before you can point the finger at one portion of your imaging system and say its a bad monitor, a bad image, a bad conversion etc.

For example lets say the you scan an image and then assign a high quality and custom profile and then convert the image to the Adobe RGB 1998 working space.

Then you convert the image to your custom made and high quality printer profile with Relative Colorimetric Rendering Intent.

Then you print the image using same as source.

The resulting print looks visually close to the original. But the image on the monitor looks blue. Rhetorically, where does the problem reside? I would look at the monitor first. I would re-profile it and triple check all of the monitors settings. My sense is that re-profiling the monitor would help. If that wasn1t it I would probably re-profile the scanner and printer and rerun my system test. I would also re check all of my Photoshop settings. Of course if I have access to more than one computer system I would try it there also.

In this scenario, I would not have been able to definitively figure that out without knowing how the complete imaging system works. Also, in this case I did not fiddle with the color I just scanned the image applied profiles and printed.

Based on my understanding of your system this is the type of methodology you need to consider to determine where this blue cast comes from.

Jim Rich


Date: Mon, 20 Jan 2003 14:47:53 -0500
From: Lippisch
Subject: Re: from an ex-pupil-sharpening, etc.

Dan Margulis wrote:

>>Similarly: if you provide me with a file you shot on a very expensive digital camera, and somebody else gives me a scan of something shot by an amateur on a disposable camera, I have certain preconceptions about how the two may differ. But once the two are open on my screen, those preconceptions are history. If the images share the same characteristics they'll be sharpened in the same way. If they don't, they'll be analyzed in the same way and each one will be sharpened differently. But where each came from is, at the point, irrelevant. They're just a couple of images that need sharpening.>>

Dan,

Point well made and taken. Once again I have learned something and really appreciate this forum! Thanks very much to you and all involved.

Alex Lippisch


Date: Mon, 20 Jan 2003 21:40:54 -0500
From: Terry Wyse
Subject: Re: Re: Color Shift on conversion to Lab

Herbert Gibson wrote:

> I have 3 test images:
> One from Andrew Rodney is ColorMatch. This converts without any
> visible change.
> A Fuji test image is sRGB - it shows a definite blue shift.
> Finally, one from Pixl in AdobeRGB converts with a very strong blue
> cast.

This sounds suspiciously like an "absolute colorimetric" conversion is going on. This would explain why ColorMatchRGB looks OK (5000K white point) and why the sRGB conversion (6500K wp) is coming up blue. Use relative or perceptual so this won't happen. Actually, you'd want to use relative going to Lab.

Terry
__________________________________
WyseConsul
Color Management Consulting
v 704.843.0858
__________________________________


Date: Tue, 21 Jan 2003 10:53:44 -0000
From: Herbert Gibson
Subject: Re: Color Shift on conversion to Lab

Terry Wyse wrote:

> This sounds suspiciously like an "absolute colorimetric" conversion is going on.

>Use relative .... going to Lab.

You had it right. Changed to Relative and all files convert perfectly now.

Many thanks.


Date: Tue, 21 Jan 2003 11:09:07 -0000
From: Stephen Marsh
Subject: Re: Color Shift on conversion to Lab

Terry wrote:

> This sounds suspiciously like an "absolute colorimetric" conversion is going
> on. This would explain why ColorMatchRGB looks OK (5000K white point) and
> why the sRGB conversion (6500K wp) is coming up blue. Use relative or
> perceptual so this won't happen. Actually, you'd want to use relative going
> to Lab.

Nice one Terry - thanks for clearing this up. I read the original post wrong in _one_ key word - I thought there was a change _of_ blue and not a change _to_ blue.<g>

Stephen Marsh.


Date: Wed, 22 Jan 2003 11:37:38 -0500
From: Stephen Jackson
Subject: Re: LAB, LAB LH and so on

>Here is an interesting message I archived from the ColorSync list from
>Florian Süssl (who I think was at Heidelberg and worked with LinoColor). It's
>some heavy duty color theory stuff but somewhat interesting in light of the
>discussion of LAB, LCH and "smaller" spaces for editing. Florina also dives
>into bit depth and the problems of editing in large colorspaces like Lab.
>While I'm sure there are those who would argue with Heidelberg that this may
>not be an issue, I'm not about to do that! These guys know their stuff.
>Shame they got out of the scanner business!

What I find strange about this post is that Heidelberg is still in the scanner business, at least that's what their website indicates, or are you referring to the individuals in the post?
Steve


Date: Wed, 22 Jan 2003 15:54:25 -0700
From: Andrew Rodney
Subject: Re: LAB, LAB LH and so on

Stephen Jackson wrote:

> What I find strange about this post is that Heidelberg is still in
> the scanner business, at least that's what their website indicates,
> or are you referring to the individuals in the post?
> Steve

They are out. The web page must need some updating. They still have to support the current crop of scanners in the market (for 7 years I think) but nothing new and a lot of people gone from the company.

Andrew Rodney

Adobe Photoshop training classes are taught in the US by Sterling Ledet & Associates, Inc.