Some new ideas from Dan Margulis

ruttrutt Registered Users Posts: 6,511 Major grins
edited April 15, 2007 in Finishing School
Dan spoke at the CS department at M.I.T. last Wednesday and I learned that he has been trying out some new ideas. I've was inspired by this and by some recent dgrin images to play with these ideas. So, here is a summary of some of those ideas and some examples (with apologies) of recent dgrin postings with some of these ideas applied.

Theory Since this was an academic talk and since Dan is a theorist, he connected his new ideas to famous color theorists.
  1. Leonardo di Vinci -- Apparently he discovered an idea that darker colors appear more neutral. At the extreme end, this is obvious, but Dan says that Leonardo meant that this loss of saturation begins much closer to the midtones than previously believed. So lighter colors should be much brighter than darker ones.
  2. Michel Eugène Chevreul -- A French color theorist from the mid 19th century was the inspiration for the impressionists. I don't pretend to understand what this work (Dan and Fredo Durand seem to agree that the standard translation into English entirely misses the point.) Essentially, though, I think that idea is that we like to see purer colors and that our eyes blend them. I'm probably doing a lot of damage to this idea, but think about impressionist (and post-impressionist) painting and this feature: using purer colors without transitions, does stand out.
Implementation
So much for theory. What about application. Dan presented a new workflow:
  1. False profile to lighten the entire image with a lower gamma, to pull the shadows into the three quarter tones and flatten in general.
  2. Convert to CMYK. Steepen the dark end of the K curve a lot to reestablish good rich shadows with better details and less saturation than in the original. Save a duplicate.
  3. Convert to LAB. Use some contrast and color enhancement technique on a layer. Overdo it. The Man from Mars technique, the standard Canyon LAB steepening technique, or the overlay layer technique from the portrait workflow are all examples of this.
  4. Use the K channel (from the saved CMYK duplicate) as a layer mask for the color and contrast enhancing step. Play with a curve on the layer mask to control how it protects the darker parts of the image from color enhancement. Flatten and now we have applied the di Vinci part theory. We have more color in the lighter parts of the image and more detail and less color in the darker parts.
  5. HIRALOAM on all three LAB channels. Do this on a layer and use a much higher opacity than you normally would. This spreads out colors and intensifies them. It's a local color enhancement just as conventional HIRALOAM is a local contrast enhancement. This gives us our Chevreul effect.
  6. Lower opacity and adjust relative lightening/brightening of this HIRALOAM enhancement. Use LAB blend-if sliders, move to RGB and use separate lighten/darken layers to adjust. Adjust opacity.
  7. Perhaps blend conventional USM with this HIRALOAM.
Examples
So how does this work out in practice? I played with two images I found on dgrin.

Here is an image by Wolfejm from a Whipping Post post.

137338889-L.jpg

I processed Dan's new workflow:

143580279-L.jpg

Without Leonardo's insight, it looks more like:

143580277-L.jpg

Perhaps a blend of these two versions would give the best realistic result?

But Dan showed something else I liked. He used the HIRALOAM layer at near full opacity and the PS brush filter to show just how close to impressionism the workflow has come.

143580318-L.jpg

In the spirit of fun, I couldn't help but try this out on another recently posted image (deepest apologies to Andy.)

143580271-L.jpg
If not now, when?

Comments

  • ruttrutt Registered Users Posts: 6,511 Major grins
    edited April 14, 2007
    Another example, this time from this WP post by gchappel.

    This time, only 3 channel HIRALOAM:

    Before:

    143381282-L.jpg

    After:

    143590767-L.jpg
    If not now, when?
  • damonffdamonff Registered Users Posts: 1,894 Major grins
    edited April 14, 2007
    VERY cool Rutt. Impressive.
  • Duffy PrattDuffy Pratt Registered Users Posts: 260 Major grins
    edited April 14, 2007
    I've got a question about the theory here. Is there something about a photographic representation that makes people start picking up on colors in the midtones that they would not see in person? When we correct a cast, its because the camera sees something that the human eye would correct for in the live scene, but cannot as easily correct for when looking at a print. This DaVinci effect seems different. It seems instead that this correction is directed toward what people tend to prefer, without regard to what they see in any particular scene. Or am I missing something?

    The impressionist technique is the same sort of thing. And its great if you are trying to give the bright, pure effect. It's definitely a better way to get the popular oversaturated effect that seems to be in vogue. But if you are going for the look of Rembrandt, e.g., you best look to other techniques.

    Duffy
  • Duffy PrattDuffy Pratt Registered Users Posts: 260 Major grins
    edited April 14, 2007
    I played with the first image some. A couple of things suggest themselves:

    1) I played some with blending the L channel at low opacity into the K channel mask. This tends to even out the transition some. I then tried playing with blending the mask into itself in overlay mode, which is probably similar to using the curves you suggested.

    2) I used the mask on the hiraloam. If shadows to mids should be more muted in color ala DaVinci, then they might also benefit by being less pure.

    Anyway, this is what I got. Its an interesting technique, with lots of room for playing around. Thanks for posting it.

    Duffy
  • ruttrutt Registered Users Posts: 6,511 Major grins
    edited April 14, 2007
    Let's see. We all agree that where there is complete black, there is no color at all. We don't see color as well in shadows as we do in midtones and highlights. But both prints and monitors aren't really black when they are trying to represent black. Black paper reflects some light and no monitor is completely black even when it's turned off, which is as dark as it can get. So we need to start getting the color out "earlier" in the lighter tones or we are going to see it when we should be seeing neutral shadows.

    I just made that up, so I don't know if it makes sense.

    As to the 3-channel HIRALOAM thing, it definitely isn't always what you want, just the way impressionism and its descendants aren't always what you want. But it's a wow when it hits and I do like the explanation. Now what exactly was it that the impressionists were seeing? I suppose that the memory of a scene isn't at all the same as the scene itself. A lot of detail is lost and color, big shapes, contrast are enhanced in memory.

    I also just made that up. I think you need to be fluent in scientific 19th century French in order to read Chevreul in the original and reason about it.
    If not now, when?
  • gchappelgchappel Registered Users Posts: 120 Major grins
    edited April 14, 2007
    Ok, I admit I am a little slow with post processing. I really like what this did to the paint-like fish image I posted to get whipped. Could you please walk me through the post process a little slower? I got lost in the workflow after the word false:D. Gamma to me is a type of radiation. LAB is where I used to work. I have read books on LAB color- but still lost wandering through the woods.
    Thanks
    Gary
    gchappel
  • ruttrutt Registered Users Posts: 6,511 Major grins
    edited April 14, 2007
    You're in luck. I didn't do very much at all to the fish image. You do need full PS, not Elements, but even a pretty old copy will do.
    1. Convert to LAB by Image->Mode->Lab Color
    2. Make a duplicate layer by Layer->Duplicate Layer
    3. Bring up the USM dialog box by Filer->Sharpen->Unsharp Mask
    4. Set the Amount to 500, Threshold to 0. Basically this means that you will see 100% of what you do. Amount is the opacity of the filter and Threshold means that it will sharpen even the smallest feature.
    5. Play with the Radius slider. The amount you need is not a constant but rather a feature of your image and of its resolution. For this effect you want to try very high amounts, maybe as high as 50 in a full resolution image. You are looking for a posterized look which emphasizes large features and intensifies colors.
    6. Now decrease amount until it looks right. People have very different ideas about this and different images call for different amounts. But it will be low, between 20 and 70, say.
    7. If you see that you've introduced some objectionable noise, try increasing the Threshold a bit, just until the noise decreases.
    8. You are done, but because you have done this on a layer you can play with the opacity a little more (and so can be a little aggressive in the Amount set above.) You can also control halos if you introduce them with the blend-if sliders in the blending modes dialog box. That's a little advanced, but worth learning. Search for "blend if" in this forum.

    gchappel wrote:
    Ok, I admit I am a little slow with post processing. I really like what this did to the paint-like fish image I posted to get whipped. Could you please walk me through the post process a little slower? I got lost in the workflow after the word false:D. Gamma to me is a type of radiation. LAB is where I used to work. I have read books on LAB color- but still lost wandering through the woods.
    Thanks
    Gary
    gchappel
    If not now, when?
  • ruttrutt Registered Users Posts: 6,511 Major grins
    edited April 14, 2007
    I played with the first image some.

    Looks good, Duffy, and I like your variations a lot.
    If not now, when?
  • gchappelgchappel Registered Users Posts: 120 Major grins
    edited April 15, 2007
    Thanks for the tutorial!!
    Gary
    gchappel
  • dandilldandill Registered Users Posts: 102 Major grins
    edited April 15, 2007
    rutt wrote:
    Another example, this time from this WP post by gchappel.

    This time, only 3 channel HIRALOAM
    I should know but do not: Do you mean HIRALOAM on the composite LAB image duplicate layer? Or, HIRALOAM done separately on each of the three channels (L, a, b) of the duplicate layer? If the former, and if only HIRALOAM is being done, is the conversion to LAB necessary?

    Hmm... HIRALOAM on a and b won't cause color shifts, just some spreading/intensification?

    Thanks
    Dan Dill

    "It is a magical time. I am reluctant to leave. Yet the shooting becomes more difficult, the path back grows black as it is without this last light. I don't do it anymore unless my husband is with me, as I am still afraid of the dark, smile.

    This was truly last light, my legs were tired, my husband could no longer read and was anxious to leave, but the magic and I, we lingered........"
    Ginger Jones
  • ruttrutt Registered Users Posts: 6,511 Major grins
    edited April 15, 2007
    Sounds like you answered your own question. Dan is using HIRALOAM on all three LAB channels at once. Experiment and maybe you'll discover something.
    dandill wrote:
    I should know but do not: Do you mean HIRALOAM on the composite LAB image duplicate layer? Or, HIRALOAM done separately on each of the three channels (L, a, b) of the duplicate layer? If the former, and if only HIRALOAM is being done, is the conversion to LAB necessary?

    Hmm... HIRALOAM on a and b won't cause color shifts, just some spreading/intensification?

    Thanks
    If not now, when?
  • Duffy PrattDuffy Pratt Registered Users Posts: 260 Major grins
    edited April 15, 2007
    When I played with this, I did the hiraloam separately for color and contrast. I don't see any reason why the contrast hiraloam should have the same values as the local color purification.

    The short answer is probably that, in most images, the result is good enough on all three channels at once. But breaking things apart will give you greater control, at the expense of some time.

    Duffy
Sign In or Register to comment.