Case study: image with wide dynamic range

Nikolai · April 5, 2007

Here's the image:

EDIT: link to original
Dan, John?

I'd really like to see what's possible to do in RGB, but if your approach leads to CMYK that's fine too.

As we discussed, it would be really great if you could explain why this area, why this channel, why this particular move, etc... :thumb

[I did a little surgery on this thread and moved the parts directly relevant to the channel vs mater curve issue to the PP5E Ch2 thread, where I think it will be easier to find in the future. Also renamed this thread appropriately-- Rutt]

Duffy Pratt · April 5, 2007

I took a stab at this one.

1) False Profile Apple 1.8
2) Curves: Held flag white, and kept dark trousers blue, steepened the red and green channels by the faces. Blue channel boosted just slightly at the midpoint.
3) Convert to LAB. Shadow/Highlights to bring out sky detail, and pants detail. Pretty heavy on both, using the technique Margulis suggests in PP5.

By the way, I think this is a poor example for showing the benefits of individual curving. There is some benefit here, but its probably very slight. The major improvements, if they are improvements, come from the false profile and the S/H adjustments.

Duffy

Nikolai · April 6, 2007

Duffy,

Duffy Pratt wrote:

I took a stab at this one.

1) False Profile Apple 1.8
2) Curves: Held flag white, and kept dark trousers blue, steepened the red and green channels by the faces. Blue channel boosted just slightly at the midpoint.
3) Convert to LAB. Shadow/Highlights to bring out sky detail, and pants detail. Pretty heavy on both, using the technique Margulis suggests in PP5.

By the way, I think this is a poor example for showing the benefits of individual curving. There is some benefit here, but its probably very slight. The major improvements, if they are improvements, come from the false profile and the S/H adjustments.

Duffy

Thank you for playing...

This is exactly the example I wanted. It's far too easy to go about different curves with a landscape, where all you have is green foliage, blue sky and red rocks. Each can be made a totaly different blue, gren and red, yet the average viewer still would not know the difference.

Here we pretty much can call every color. Everyone know what does the flag look like. Most people know the color schema of the federate army. Shot has dual lighting - sun and the fill flash. It *is* a real situation I have to deal with.

Now, with all due respect....

Your treatment made the shot worse than it was.

One thing Mr. Margulis keeps saying acrros the board is "do not give the viewers the color they would know better than to believe".

Union trousers cannot be green. North army was not called "Blue" for nothing...

Shadows cannot be green either, not in the plain field, reddish ground and the blue troops casting it. If nothing else, I was there myself and I do not recall any green. Even my 30D does not remember one.

Yet your treatment shows A(10) in shadows and A(17) in pants.

Sorry, no hit...

Duffy Pratt · April 6, 2007

The pants in the original are already negative in both the A and B. You have a point about he shadows. However, the faces are strongly magenta in the original shot.

I took another shot at this. This time, no false profile. Instead I first created a new layer, luminosity mode, and applied the image to itself in Screen mode at about 50%.

Then curves as before, but this time I locked down the shadow and pants area before trying to increase the contrast in the faces and getting the skin tones a little better. Again, I locked down the white in the flag for a highlight in each channel.

And the same shadow/highlight adjustment as before in LAB.

Also, keep in mind that there is a difference between the effectiveness of a technique in principle, and the effectiveness of it in my hands. I'm still learning, and I never paid any attention to the shadows. To a cerain extent, I believe that if someone is looking closely at that shadow, then the picture is a failure already. I appreciate the feedback, even if it is a thumbs down overall, because I'm still learning this stuff (only been at PS for a little less than a year now).

Duffy

Nikolai · April 6, 2007

Duffy,

Duffy Pratt wrote:

The pants in the original are already negative in both the A and B. You have a point about he shadows. However, the faces are strongly magenta in the original shot.

I took another shot at this. This time, no false profile. Instead I first created a new layer, luminosity mode, and applied the image to itself in Screen mode at about 50%.

Then curves as before, but this time I locked down the shadow and pants area before trying to increase the contrast in the faces and getting the skin tones a little better. Again, I locked down the white in the flag for a highlight in each channel.

And the same shadow/highlight adjustment as before in LAB.

Also, keep in mind that there is a difference between the effectiveness of a technique in principle, and the effectiveness of it in my hands. I'm still learning, and I never paid any attention to the shadows. To a cerain extent, I believe that if someone is looking closely at that shadow, then the picture is a failure already. I appreciate the feedback, even if it is a thumbs down overall, because I'm still learning this stuff (only been at PS for a little less than a year now).

Duffy

Once again, thank you for your time!

When you say "locked" what do you mean? Selection/mask? Isn't it a big no-no in PP5E?

I'm getting a green tint on the gound now (and on lady's apron)...

I guess I need to put a bigger version of file online, cause it can make a huge difference in perception..

And, just to clear things out: we're not trying to simply improve the shot in every possible way. It was supposed to be an exercise in exactly channel-based curves, since I, for one, still can't understand how Dan (Margulis) is doing it on a channel by channel basis without selection/mask and without screwing other parts of image.

I'm starting to get a suspicion that his technique is, in fact, only possible when an image contains
1) only subset of color palette (any landscape can be a good example), or
2) colors that can be tweaked without being caught redhanded (landscape again, since it can't examine its one picture and say "hey, my leaves are not that brownish...":-)

I will locate the original and add a link to a larger file tonight.

Thank you again!

Duffy Pratt · April 6, 2007

When you say "locked" what do you mean? Selection/mask? Isn't it a big no-no in PP5E?

By locking, I put a point in the curve (on all three channels) at the top range of the blue pants, and then somewhere below it in the range of the pants. This keeps that part of the curve straight. Since the faces are higher up on the curve, they can still be adjusted.

I'm getting a green tint on the gound now (and on lady's apron)...

The ladies apron is slightly A negative in the original. What color is it really? If you are willing to bet the picture on its being white, that might be helpful.

What's clear to me is that the original is not a fair representation of the colors. The sky is definitely off, as are the faces. What we are looking for is a good improvement in the time that someone is willing to spend on it. With a limited amount of time, and my limited abilities, something is going to give. But I still think that both of my versions, and certainly the second one, are improvements over the original.

And the ground has gone greenish yellow now. In context, does it look obviously wrong? What do you prefer, a slightly green grown, or strongly magenta faces?

I'll take another look.

As for the value of channel-based curves, the question is whether it is better than using the master curve. Can someone make a better improvement to this picture by simply using the techniques I have and the master RGB curve?

Duffy

Duffy Pratt · April 6, 2007

OK, I took a third shot at it.

This time we go to the same steps as in two.

The RGB curves have points at the following areas:
Red: 0,0, 32,28, 137,174, 240,240 and 255,255
Green: 0,0, 62,61, 75,75, 139,174, 234,237, and 255,255
Blue: 0,0, 64,64, 87,87, 119,111, and 255,255.

I did the curve on a dupe layer instead of an adjustment layer.

Then I moved to LAB, and with the blend if sliders, I made it so the top layer would not have any effect in areas that were A negative on the top layer.

Then I did an adjustment curves layer to make the sky bluer. This was a curve that killed the A negative and steepened the B negative. Blend if sliders limited this adjustment to where the underlying layer was in the highlights and was also A negative.

Then I did the Shadow/Highlights as before.

Duffy

jfriend · April 7, 2007

rutt wrote:

Duffy, I've also been thinking about the reaction to those early chapters, especially in comparison to the success of the LAB book. I've even talked to Dan about it. Of course, these days, he's experimenting with something totally different: false profile to decrease gamma a lot, convert to CMYK. Steepen K, keep copy of K channel somewhere, convert to LAB. Steepen A+B through a layer mask containing that saved K channel, HIRALOAM all three channels. Of course, his results look great, mine, not so much yet. Dan demoed this at MIT on Wednesday.

In the last advanced class I took, Dan really stressed luminosity blending for contrast enhancement. I'm just beginning to get my mind around this.

If I were Dan (or his editor), I'd unify PP and the LAB book. I'd start with the first four chapters of the LAB book. I'd do impossible colors, including RGB channel curves for color. I'd teach blending. I'd teach sharpening including HIRALOAM and how to blend HIRALOAM and USM. Only then would I get into RGB channel curves for contrast.

But then who am I? I can't reproduce Dan's results even after they have been clearly demonstrated any number of times...

This is a bit off topic, but I was just playing around with this image last night (didn't save any of my changes) and discovered that this image really, really lends itself to a steepending of the K curve in CMYK because it really helps contast in the shadows without messing with the rest of the image and without messing with the color and without impacting the highlights. Since I still rarely use CMYK, this was a cool discovery for me and worth some others trying too.

Beyond that new discovery for me, I certainly find this image a lot easier to process in the traditional methods (separating color from contrast).

rutt · April 7, 2007

Yes, we are off topic. That's because Nik is expecting the impossible. Not every image benefits from every technique. We can't expect that just any old image that someone happens to dig up will be amenable to a particular technique. In this case, first Andy and then Nik have posted some image or another and then posed a challenge to improve with channel based RGB curves. Much better to show images where this technique does make hay. Then keep it in mind for the time when it solves the problem at hand. As I've been saying, I don't think channel based curves for contrast is the most widely applicable technique that Dan teaches and it's too bad it gets introduced so early in PP5E. I think that's a bit of a historical accident. In previous editions, cast removal comes first and for that problem, channel based curves are a very valuable technique.

In the case of this particular image and most portraits, using channel based curves for contrast enhancement is like using your sand wedge for teeing off on a par 5 hole.

Enough hot air. I gave this image a no-hold-barred attempt and got this result:

What did I do?

In RGB: Lots of luminosity blending to bring out the sky, faces and some of the shadow detail. The blue channel turned out to be more useful than I would have guessed to darken the flat skin and to lighten the uniforms a bit. The red channel was great for the sky and for the lighter colored pants (to keep them from getting too light.) Used channel based curves to get some more yellow in the faces.

In CMYK: Steepen the K curve a lot (hey channel-based curves)

In LAB: Command-click curves to get better reds and blues in the flag, add more blue and less green to the sky. Deepen the blue of the uniforms. Use the K channel from the previous step as a layer mask for this this curve layer, which has the effect of focusing the color enhancements on the highlights and midtones and keeping them away from the darker parts fo the image (ala Dan's new techniques.) HIRALOAM on all three channels with a very light touch (about 25%) and about 1/2 as much lightening as darkening. Conventional sharpening on a separate layer at about 50% with blend-if to allow the HIRALOAM shadows through.

Looks darn good to me, but I worked on it way too long. Of course I was practicing and learning some new techniques as well as trying to clear up my thoughts about channel-based-curves.

So?

rutt · April 7, 2007

Better, but I'm still not happy with it. I want better shadow detail in those uniforms but I don't want it to look fake. They can't be too blue.

Nik, will you make the raw version available. I want more resolution in the starting point.

Way way off topic, but jeez I think this is kind of an interesting challenge as an image.

Duffy Pratt · April 7, 2007

The challenge in this image is that it is so unlike most images in terms of what you think you know. Everyone thinks they know the color of the uniforms, the skin, the sky, the dirt to some extent, the flag, the sabres, the gunstocks, etc... Color correction usually involves some trade-offs. You nail the known areas and let the unknown areas slip somewhat. Here, you pretty much can't do that.

Compound that with the fact that some of the "known" values in this picture may not be true: How new is the flag? How much have the colors faded? Should the picture account for the fading or not. Is the apron white or something else? Are the trousers really blue, or closer to some sort of cyan? What happens when you apply some red dirt and dust to the trousers -- should they go to grey, or somewhere else?

And those are only the problems on the color side. The starting exposure on the troops brings up a host of contrast problems as well. And the amazing thing is that all of these problems arise in a shot that is pretty well done to begin with.

Duffy

Nikolai · April 7, 2007

Link to original
http://nik.smugmug.com/photos/141824921-O.jpg

rutt · April 7, 2007

Nikolai wrote:

http://nik.smugmug.com/photos/141824921-O.jpg

No raw, Nik?

Nikolai · April 7, 2007

John,

rutt wrote:

No raw, Nik?

SM does not take raws...:cry
Hold on, I'll post one on S*E site:-) Stay tuned!
....
Here you are

Andy · April 7, 2007

Nikolai wrote:

SM does not take raws...:cry

Do so!

http://smugmug.jot.com/WikiHome/PhotoshopSampleFiles

Nikolai · April 7, 2007

rutt wrote:

That's because Nik is expecting the impossible.

I'm not:-) I only want to see how this stuff is played outside the Canyon Conundrum :-)

rutt wrote:

So?

What a great processing (with the exception that don't like the halos on the closest trooper)!

Nikolai · April 7, 2007

Duffy,

Duffy Pratt wrote:

The challenge in this image is that it is so unlike most images in terms of what you think you know. Everyone thinks they know the color of the uniforms, the skin, the sky, the dirt to some extent, the flag, the sabres, the gunstocks, etc... Color correction usually involves some trade-offs. You nail the known areas and let the unknown areas slip somewhat. Here, you pretty much can't do that.

Compound that with the fact that some of the "known" values in this picture may not be true: How new is the flag? How much have the colors faded? Should the picture account for the fading or not. Is the apron white or something else? Are the trousers really blue, or closer to some sort of cyan? What happens when you apply some red dirt and dust to the trousers -- should they go to grey, or somewhere else?

And those are only the problems on the color side. The starting exposure on the troops brings up a host of contrast problems as well. And the amazing thing is that all of these problems arise in a shot that is pretty well done to begin with.

Duffy

Hey, that's exactly why I insisted on bringing in my own image

I understand that this is probably the least preferable candidate for the said technique, but this is why I wanted to see how it can be done properly, if at all...

Nikolai · April 7, 2007

Andy,

Andy wrote:

Do so!

http://smugmug.jot.com/WikiHome/PhotoshopSampleFiles

I guess I missed something...

I clicked on the Upload document, and it told me I don't have a permission. :cry
Can you please point me towards more info on the subject?

Andy · April 7, 2007

Nikolai wrote:

I guess I missed something...
I clicked on the Upload document, and it told me I don't have a permission. :cry
Can you please point me towards more info on the subject?

Are you registered on the wiki

Nikolai · April 7, 2007

Andy,

Andy wrote:

Are you registered on the wiki

I did just recently, but here's what have been getting from the jogspot for quite some time:

New users

We've closed off new account registrations while we focus on migrating to Google's systems. If you'd like to be notified when we re-open registration, enter your email address below.

I guess I'll be SOL for a while

Andy · April 7, 2007

Nikolai wrote:

I did just recently, but here's what have been getting from the jogspot for quite some time:

I guess I'll be SOL for a while

Oops. Forgot.
put it on your server, for now, Nik

Nikolai · April 7, 2007

Andy wrote:

Oops. Forgot.
put it on your server, for now, Nik

I did :-) http://www.starexplorer.com/download/IMG_9521.zip

Do I understand it correctly you can only do through wiki, not via regular SM/API?

rutt · April 7, 2007

Andy wrote:

Rutt, why are the pants of the guy just to our right of the pipe-smoker, dirty in your shot?

Because I was guessing that they shouldn't be as blue as you got them.

I'm still waiting for a raw version and I'll try again, I guess.

Andy · April 7, 2007

Nikolai wrote:

I did :-) http://www.starexplorer.com/download/IMG_9521.zip

Do I understand it correctly you can only do through wiki, not via regular SM/API?

No raw on SmugMug, yes RAW on our Wiki for these purposes.

Nikolai · April 7, 2007

Rutt,

rutt wrote:

I'm still waiting for a raw version and I'll try again, I guess.

It's been uploaded to S*E site, please check my recent posts

Nikolai · April 7, 2007

Andy,

Andy wrote:

New i2e version, from Nik's raw file.

Very, VERY, VERY impressive!

What was that deal on i2e s/w again?

Duffy Pratt · April 8, 2007

OK, I think I might have learned something from this thread. Here is another attempt. I did too much to describe in detail. It included:

1) False Profile 1.4 gamma.
2) Luminosity blends, Red into RGB, Blue into Green. Both limited by blend ifs.
3) Steepening K and Magenta channels in CMYK on Luminosity layer.
4) Color balancing skin tones in CMYK color layer.
5) Lab for a shadows move, and to play some with the blue in the pants and sky, and an unsharp mask.

I think this is better than my other efforts, and I still like this picture lighter than others have made it.

Duffy

Nikolai · April 8, 2007

Duffy,

Duffy Pratt wrote:

OK, I think I might have learned something from this thread. Here is another attempt. I did too much to describe in detail. It included:

1) False Profile 1.4 gamma.
2) Luminosity blends, Red into RGB, Blue into Green. Both limited by blend ifs.
3) Steepening K and Magenta channels in CMYK on Luminosity layer.
4) Color balancing skin tones in CMYK color layer.
5) Lab for a shadows move, and to play some with the blue in the pants and sky, and an unsharp mask.

I think this is better than my other efforts, and I still like this picture lighter than others have made it.

Duffy

I think this is definitely better than the previous tries

Yes, you made it lighter, I guess it's a judgement call. I should mention that this version delivers the feeling of the heat downing upon the marching troops from the merciless sun...

Duffy Pratt · April 8, 2007

One more general observation about this image. Because the color areas are so "uniform," it is really quite easy to do what amounts to fake selections using the "blend if" sliders. The deep blue jackets lie in one area, the light trousers in another, the sky has its own unique aspects, as does the skin. In many ways, this is a very non-typical image.

Duffy

Nikolai · April 8, 2007

Dyffy

Duffy Pratt wrote:

In many ways, this is a very non-typical image.
Duffy

Yet this is of the kind that gets purchased, so it's quite important from all practical purposes.

Reenactments are my best sellers, if you don't count appointments...

rutt · April 8, 2007

Duffy, much closer to what I was looking for. Unplugging the shadows in those uniforms makes a huge difference to me. I have an idea I want to try out, but it's not that different from what you described. HIRALOAM the K channel and stamp it into the false profiled (and maybe shadow/highlighted) RGB with a multiply layer.

Case study: image with wide dynamic range

Comments