Courtyard Wash: Dan's comments on individual versions

Dan Margulis

The numbers by each entry were sampled (sample size 51x51) at a point in the light area in the center of the lower wall, intended to represent the lightest natural rock; and also in the chest of the secondary figure. The readings are not as valuable as I thought they might be, but here is some guidance.
*Some people, as in 1101, have the secondary figure quite light due to blending choices, so their numbers for that area shouldn’t be compared with those of others. That said, there should be at least 30L difference between the two measurements. The greater the difference, the more contrast between figures and background. The better versions seem to have the difference at around 40L or even more. Again, not a hard and fast rule.
*This one is hard and fast. The first measurement, the light part of the wall, needs to be a warm yellow or an orange. On reviewing these entries I suggest that 0 < 1.5A < B < 6A are the desired values and if they aren’t met the image isn’t adequate.
*The second value should also not be too colorful. If the secondary figure is perfectly neutral then it’s 0A0B, but it should probably be warmer, so both numbers should be positive. My first recommendation was A+B < 25 as otherwise things would be too colorful. On further review that seems to conservative. I’d say A+B < 30 is more reasonable.
*Also, that second color needs to be somewhat accurate. It must be warm, meaning neither the A nor B can be negative. In principle the B has to be greater than the A, otherwise the glyph is on the magenta rather than the orange side of red. It does seem beneficial artistically here, however, to have the measurement differ from that of the wall, preferably by having the glyph more neutral, but possibly also by having it more magenta. So I said that if A > B+8 there is a problem. I guess I stick to that recommendation.
Below I’ve outlined the demonstrations that created the four alternate par versions that I posted yesterday. The links are in my previous message.
1101 (91L 3a 24b; 68L 9a 11b) Chosen for the par version, not because I particularly like it but because it represents the simplest approach: highlight the glyphs by making the background light and uninteresting.
Robert Wheeler’s approach was not particularly simple. He concluded that he could not get the detail he wanted out of the original shot, so he would derive it exclusively from the infrareds. He converted them all to LAB, discarded the ones that were obviously unsuitable, did an Auto on the L channel and converted to grayscale. Then, he stacked them up and did various multiplications to get what he wanted. Finally, he put his result on top of the original in Luminosity mode. After merging, he increased saturation slightly.
Note that by the numbers the images are strongly yellow, yet we perceive them as black.
1102 (71L 16a 36b; 35L 12a 11b) Ken Harris, who posted a complete description, did not use the IR. He had a copy of DStretch, which is software specifically designed to extract images from rock art, making several different versions with it and then combining them. He admits that the color isn’t accurate, but given the nature of the restoration the added warmth is certainly an acceptable exaggeration and falls within my recommendations above.
The real problem, though, is, that in spite of all the good work resurrecting the glyphs, the version is improved by running Auto Tone. A demonstration involving this version at #1103, and another comment at #1110.
1103 (80L 10a 27b; 40L 33a 26b) Chosen for the par version, but #1102 would have been a better choice. Like Ken, Ed Bateman basically made no use of the IR.  Like many of us, he made a couple of different versions for luminosity and an additional one for color only. Here are his notes: 
My first two luminosity versions:
Version 1: I made one (IMG0052) where I used Equalize on each channel. Then I duplicated that layer - inverted it - and Gaussian blurred it at 108 blending with soft Light to tame the contrast.
Version 2: The next version started with (IMG0052) Green channel and curves. The I duplicated that layer, Gaussian blurred it and and blended with Overlay.
COLOR: I made a Color boosted layer using Man From Mars in LAB. (Because I didn’t know where it would take me…) I added my version 2 above that using Luminosity. I added version 1 on top of that using Pin light - at 95% opacity.  This became my color boost layer.
On My Master image (in LAB), I copied Version 1 into it with a blending mode of Luminosity and used default Shadows/Highlights. On top of that, I added Version two with a blending mode of 48% - also using shadows highlights.
I added my Color master to this and used some Color Blend-if adjustments. I next made a composition layer and used Replace Color to change a few colors on the glyphs that had become a bit vibrant and funky. I did a bit of saturation adjustments on the top and bottom to keep them within reasonable saturation. And a bit of last minute curve for color.
Finally, on a copy, I did a Highpass Overlay move - applied it using Darken to only add some contrast to the dark areas and added that to the image.
All this provides a very snappy image, which is why I initially preferred it. I now find #1102 more convincing, however, because it lacks some of the problems that this version has.
Ed should have artificially lightened the white-pigmented areas as Ken did. But the real issue is color. According to the measurements the light area of the wall is good. It’s a little more vivid than most others but that makes for a happier appearance.
The glyphs, though, show up as A+B=59 (vs. A+B=23 in #1102). This is well over my recommendation that the number should be less than 30, and indicates they are excessively colorful.
MMM should not have been used here. It doesn’t help the glyphs, in fact it makes them redder, and adding color variation to the wall distracts us from the subject of the image. Also the Color Boost layer should have been masked with something that reduced the effect as the image gets darker. The bright colors in the light half of the image aren’t bad but the entire upper right half is a problem.
Still, the excellent tonal contrast between the wall and the glyphs could be used to improve #1102, with something like this:
Ingredients: #1102 and #1103
1) Start with #1102 and add a duplicate layer.
2) To it, apply #1103 in Lighten mode, which is needed to prevent the white areas of #1102 from getting darker.
3) Change layer mode to Luminosity to restore the color of #1102.
4) Add as a layer mask, a copy of the red channel, so as to protect the darker parts of #1102 while allowing the lighter parts of the all to get lighter.
5) Reduce layer opacity to taste. I chose 50%.
6) Duplicate the layer (and mask), change mode from Luminosity to Color, and decide whether to use any of it. I could see doing this at a very low opacity, limited to the light areas, but thought it too much bother and threw the layer away.
7) Flatten image and apply Auto Tone.
The result is #1118, which according to me is better than the par, as well as either parent.
1104 (81L 5a 3b; 63L 10a (5)b) The numbers indicate that the whole image has a magenta cast, but the real issue is lack of definition in the figures.
1105 (90L 5a 4b; 56L 11a 14b) This person’s process notes state that there was a lot of experimentation but that most of the detail came from IR version #47 and adds “I am more interested in glyph definition than naturalistic color.”
There is some truth to this, but MMM+CB should not have been used, there is no reason to have the right side as cool as it is, it looks more like it was shot in a cave.
1106 (90L 3a 9b; 49L 31a 15b) This individual is one of the few who dredged detail out of the A and B channels of the IR. These have much more variation in the figures than the luminosity-based channels do, so it’s a good way to create a more mysterious look, where the images are faded non-uniformly.  In principle, I like the detailing of his main figure better than many others.
The color is another story. He says that his understanding of these ancient paintings is that the colorants were typically ochre or brown, when they were not black. And he mentions that blood may have been used as a colorant, which could account for the magenta flavor of the secondary figure.
Even accepting this, the numbers confirm the obvious, that this version is too colorful, especially the wild yellows. If a yellowish feeling is desired, I’d suggest the following.
Ingredients: #1106 only.
1) In Hue/Saturation, reduce master saturation -40.
2) Duplicate Layer
3) Apply the Camera Raw Dehaze filter at 100%, to increase the perception of the shapes.
4) As this move plugs the entire upper right, add as a layer mask a copy of the RGB composite.
5) Blur the layer mask 30.0 pixels.
The result is posted as #1119.
1107 (85L (7)a 10b; 57L 0a (2)b) Too much coolness everywhere, particularly in the foreground rock, not enough contrast on the left side, upper right is plugged.
1108 (81L 8a 15b; 50L 16a 1b) Harvey Nagai reviewed all the links, seeking inspiration, and then a few more that I hadn’t listed. He went through the familiar litany of version after version, trying to strike gold, and instead struck magenta, which I thought highly objectionable then but not so much now.
I like it better now, it’s a nice conservative treatment. Comparing it to the par shows why the figures shouldn’t be as magenta as Harvey has them. But the comparison also shows up some problems in the par that Harvey does not have, wherefore
Ingredients: #1108 and the par, #1117.
1) Start with #1108 and add two duplicate layers. Make the top one invisible.
2) To the middle layer, apply the par in Lighten mode. This mode protects the upper right, where Harvey’s version is the better of the two.
3) Activate the top layer, a copy of #1108. Change mode to Color and adjust opacity to taste. I chose 50%.
The result is posted as #1122. The reason for the last step: I find the par monotonously brown. Harvey’s original color is likely too much of a good thing, but blending the two works for me. Another comment about this version is at #1116.
1109 (84L 7a 37b; 44L 22a 28b) Here’s another that I like better now than I did then. John Furnes:
This has been the most interesting exercise / challenge of all. And the most enjoyable.
I have not managed to render it “as it possibly was originally”. And I have not painted in or added colour other that what have emerged from the originals and the treatment in PS.
Most of my trials have been manipulating and blending of layers.
I think I have done some thirty five or so tries.
Eventually I found that only a few of the pictures matched size and position. Hence, I opted for those, and started anew.
After having tried all I know about Photoshop, I chose two images to start out: (so the extra time was very welcome)
42GR00605_080528_IMG0044 and 42GR00605_080528_IMG0045 (IR)
1 No. 45 I did a few adjustments in ACR, and
2 then into PS to make it a BW.
3 To emphasize contrast, I went to LAB, and neutralised a and b (with curves)= grey
4 Then I applied High Pass to L (at 30)
5 Opened No. 44 in ACR.
6 Then to Lab
7 Adjusted as much as possible to emphasize the different forms, and tried to get more contrast without losing texture
8 Color Boost 2018 Lab: this brought out different colours in the pictograms, and gave new hope
9 Copied two times, and layered as
10 Saturation
11 Color
12 Copied in No.45 (now as BW) as Luminosity at 20%
13 Tweaked  quite a bit the different layers, color was switched into Luminosity, and some Layer Masks were manipulated (Apply Image L at Normal). Color Dodge with Fill at 23%, and so forth.
14 Sharpen 2018: adjusted several layers
15 sRGB
16 Save
I think I might have skipped quite a few of the steps, as I tried almost every move I could think of. I have never blended so much for one single image.
I did not think that the IR-pictures were much help – they did not give much contrast or definition. Still, I used them for making just that – contrast as BW.
I look forward to see the rest of the guys’ tries, and hopefully you have one too.
The detailing is good but as the numbers indicate, it’s both too yellow and too colorful. To show that there’s merit in it, I find that blending 25% of #1109 into the par improves it both for color and detail.
1110 (86L 3a 18b; 48L 4a 0b) Having seen a couple of versions that I like better than the first time around, here’s one that I like worse. Regrettably, it’s my own. 
The good things about it: I prefer the weathered look of the left side graphics to the crisper renditions elsewhere. I did not bother to align the LAB images I was working with since I reasoned that a little blurriness is not a bad thing for artwork this old.
Also, from my point of view this color is the most accurate to the scene of any version except #1116. That isn’t to say it’s the best color for this purpose, only that many of the others are too brown or have overly colorful figures.
This was a first effort and I wasn’t very happy with it although I couldn’t put my finger on why.  So I did a second version this weekend but it was even worse and I trashed it.
On reviewing the work of others I saw what the problem was. If you take any of the more orange IRs into LAB and apply Auto Tone or similar to the B channel, a large amount of usable detail pops out. So does a lot of noise. In the figures themselves, this is useful. It gives the weathered appearance that I’m looking for. But in the rock face it’s a bad thing. If the rock was as rough as I portray it in #1110 then the Fremont people would never have tried to paint on it.
The solution (which a couple of other people could try): in my first effort I used the dodge tool to lighten the white-pigmented areas, and also to lighten the area between the horns of the two main figures. I should have realized that if I was going to do this kind of retouching I might as well get the whole wall as well. And since everybody else was wasting a lot of time on this image I decided to do a new version post-mortem. So,
DEMONSTRATION, skipping over lots of steps:
Ingredients: Raw versions 44 and 52.
1) I took the IR version into LAB, opened the contrast in all channels, then blended them together as a grayscale in a way that hopefully preserved the weather look.
2) From the original raw, I applied curves to neutralize the figures and achieve a version correct for color.
3) I then merged the contrast of the grayscale version with the corrected color.
4) Now, duplicate the layer of the resulting image. It is full of the kind of noise in the wall that is found in #1110.
5) Apply Surface Blur, Radius 30, Threshold 45, wiping out almost all the noise.
6) Add a layer mask and apply the original red channel, which has the best contrast between wall and figures.
7) To this mask, Auto Tone.
8) Reduce layer opacity to taste. Since we don’t want the wall to look like it’s been sanded down to complete smoothness, I chose 50%.
9) Flatten and do the necessary retouching to the light areas.
The result is posted as #1120. It’s about what I was intending to get in #1110. I like it better than the par, but something even better is about to pop up.
1111 (82L (1)a 3b; 35L 18a (13)b) This person’s process notes:
- analyzed each color image and forensically determined what colors should be, in what locations (mostly symbols/glyphs); make a sketch to track all parts of the image
- Assessed all source images (and searched for more) and their channels and make selections as needed to find/make masks for every part of the image from various RGB, Lab, CMYK
- Use/make the best base image and add layers with masks to manage and assemble all parts of the image to yield the most correct colors and luminance for all parts of the image; where masks are not available or imperfect, manually"paint" to make/adjust masks
- Use a variety of adjustment layers with masks to get desired results; such as H/S, fill color, etc and use blending modes as needed to yield most correct colors, contrast, luminosity, etc.
First result seemed too yellow, made second which was too red; blended them for final.
All this planning paid off, with clearly the best shape of any of the 17 posted versions. If you don’t believe it, put it as a duplicate layer on top of your own version and set it to Luminosity mode.
Unfortunately, the color is not usable. The psychedelic purples of the main figures are impossible. We need a blending partner. If have just declared that this one has the best contrast of any and that #1110 has the most accurate color, so that’s a natural marriage. Furthermore, we can also address the complaint that the color of #1110 is, though accurate, boring.
Ingredients: #1111 and #1110.
1) Start with #1110 and make a new layer.
2) To it, apply #1111.
3) Change mode to Luminosity.
4) Evaluate whether to reduce opacity. I found the effect slightly too strong and reduced opacity to 90%.
5) Flatten and find some way to boost color in the lighter half of the image. I moved it into LAB and applied the Color Boost 2008 action. I masked it with a copy of the L channel, and cut the overall opacity of the action in half.
The result is #1121, and as you might expect, it’s my favorite of all those posted.
1112 (80L 7a 9b; 44L 16a 11b) This one measures too brown, not yellow enough. It could also use a luminosity blend of the red channel into the composite, to add definition to the figures.
1113 (76L 18a 44b; 28L 24a 20b) Chosen for the par version. The overall color is similar to that of #1102 and #1103, but is extremely intense, too colorful by my by-the-numbers standard, as though the thing is artificially lit in a museum. I’m fine with that, since the viewers are fully aware that this is an artificial restoration. This person’s process notes:
So... took the default 0044.CR2 and Lab'd up the saturation, took the infrared 0045.CR2 - in Lightroom, de-saturated in basic panel & set white balance to 2,000, tint to -150, to get a more contrasty grayscale rendering—then used Dehaze at +100 to really bump the contrast, set white slider to +58, and black to -17 —blended the Lab color into the grayscale and adjusted Curves and Hue/Sat to get something reasonable. The hardest thing about this is deciding how much contrast and color to apply. A birghtening Curve was masked-into areas that would have been white, and white edge halos were added to more or less match the Pre-1980 Moab-Panel jpg.Tried not to obscure anything that was in the infrared file, even if it didn't match.
Spent about 15 minutes on this... too busy for more.
1114 (85L 5a 21b; 38L 10a 5b) Chosen for the par version. The color is similar to that of #1110, so the demonstration there of blending with #1111 would work here as well. This person used a gradient mask to protect the shadow area, but should have protected the whiteness above the figures so that it wouldn’t be so distracting. Dodging between the horns of the two main figures would also have helped. This person I think was the only one noting the magenta cast in the main figures as being incorrect in principle, but that analysis of the captures indicated that the purple color was natural and not a camera WB issue.  The process notes:
Several things made this a challenge.

1)  All images were misaligned to varying degree
2)  Lighting changed among different exposures (hard shadow, soft shadow)
3)  The obvious lack of color info, and the unknown target

I landed in a few different places, and could see a case for two primary things, one withe more purple, the other with more brown.  The question for me is if the purple was a byproduct of the WB lighting or not.  Looking deep into the image 200%, it looks like I could see variation that would suggest it was not consistent to a WB issue, so I made the assumption that the purple was a viable color, in concert with the browns / blacks.  But, a less colorful version (no purple) would certainly be viable too.  Trying to conceive what "might have been" vs. what we know after time has taken its toll opens up a realm that is philosophically in nature.  Rambling aside, here's my process.

1 Taken original color file into ACR and process as a monotone to set tonal values as I want them (save this for later).
2 Take all 9 files, color +  8 of the IR files and use photomerge to create a stack layer that aligns the layers.  This created a file slightly larger than the original size by about 10 pixels each direction.  (later to crop back down to the correct file size at the OP color image).
3 Set the base color file as the background layer, and each of the subsequent IR images, set to MULTIPLY blending mode.  As there was 8 layers, I weighted them at 88%.  Sounds arbitray, but it is actually 1/8 = 12%, 100% - 12% = 88%, so I only took away a little of each, rather than only leaving a little of each to contribute.
4 The stacked layers on multiply created a virtually black image.
5 Using the monochrome file previously from step one, I used it as a layer set to SCREEN to restore luminosity.  Actually, this was two successive layers.
6 After this, there was a strong dark shadow that prevented us from seeing the detail in the upper right section.  From there, it was a decision to use gradient mask / painting mask to bring the shadow areas to a more visible state.  Since this is an exercise to find out what might have been, I gave less credence to the "as shot" shadow.  The point was to undertake a CSI recreation / restoration effort ... so, I genuinely wanted to know / understand what was in there ... shadow cast be darned.
7 After that, mostly tweaks tonality, gamma / sat to render plausibly vibrant (we don't know) vs. excessive.
8 In the end, I took the OP color file and set it to hue, and adjusted opacity down to take just a bit of vibrancy back.  How vibrant to render is probably the most ??? part, comparing the what might have been vs. the toll of time.  
NOTE:  In the process I wound up with an artifact at the bottom of the frame (misalignment stuff), that I couldn't really get rid of well, without redoing the entire process.  But, for some reason, when I went to repeat the process (willing to do so), my photomerge would consolidate the 8 IR files into only two layers, one orange, one purple ... weird.  So, I just putzed with the OP overlaying the artifact.  Bummer, but it wasn't the real mission, since it didn't impact the drawing area.  So, if you look at the bottom edge and it looks funky ... thats' why.  I might try again ... but, with the deadline here ... well, here it is.
1115 (92L (1)a 2b; 57L 5a (5)b) This person used a variety of filters, as well as 12 layers and a lot of alpha channels. The result is quite good on the right side of the picture but the whole left half appears blown out by comparison, and having it neutral doesn’t help.
1116 (80L 5a 11b; 50L 17a 1b) Chosen for the par version, but I can’t imagine why I picked it instead of 1108, which has the same color without the distracting red roof and without being oversharpened. The oversharpening can be corrected with the same technique as shown in #1110.
1117 (82L 8a 25b; 45L 18a 12b) The par version.
1118 (82L 12a 13b; 38L 9a 3b) The alternate par created at #1103.
1119 (89L 2a 5b; 41L 23a 6b) The alternate par created at #1106.
1120 (90L 2a 12b; 46L 6a (3)b) The alternate par created at #1110.
1121 (83L 4a 23b; 37L 5a 0b) The alternate par created at #1111.
1122 (84L 8a 18b; 51L 17a 2b) The alternate par created at #1108.

Join to automatically receive all group messages.