Panama 1978: comments on individual versions

Dan Margulis

Here are my notes on this set. Where names are used, the person has already identified himself in a post to the list.

701 The yellowish feeling is deliberate on the part of Gerald Bakker, who calls this one the most difficult case study so far. He blurred the AB channels and used Dust & Scratches on the L to try to eliminate the pattern. The color overall is conservative yet reasonable, but the overall impression lacks depth. As noted in one of my posts, the easy way to fix this, even now, is by blending the green into the red on a layer set to Luminosity. It is important that this blend be done in Darken rather than Normal mode, as otherwise all the blue clothing will be damaged.

It should also be noted that, like the Cinque Terre study, this one benefits from H-K, and for the same reason. Forgetting the green parakeet and the little girl’s pink color, which are too small to matter, there are plenty of different colors here but nothing particularly intense. In a situation like that H-K is useful to create more variation. Here, it also has the benefit of darkening the wall, which creates an attractive contrast with the faces. Even with this completed image in its current state, it would still be beneficial to run H-K followed by some kind of color boost to restore the overall feeling.

702 An excellent presentation but, like the similar #713, no effort to cut the pattern. Otherwise quite realistic.

703 Every now and then we need a reminder of why we need a full tonal range every time. This person used a fairly simple process to establish vibrant but believable color and then wrecked it by not establishing a valid white point. At this point running Auto Contrast changes the image from one of the worst entrants to one of the best.

704 The contrast/detailing is good, the result of several blends plus H-K and the Lesser Hammer. The blurring is somewhat overdone. The overall impression is marginally too gray. An easy way to change that impression is to use the Sponge tool to saturate the parakeet and the little girl’s magenta collar. Such small flashes of brightness help convince the viewer that he’s not looking at a grayscale image, and we don’t have to worry about holding detail in things this small.

705 Similar to #704 in being basically correct for color but somewhat gray, with reasonable detail. The person writes, “I found it quite a challenge to get good detail and contrast in this image and to manage the tradeoff between sharpness and noise so it will be interesting to see the group results.” He used that Pattern Suppression plugin, followed by a light application of Dust & Scratches and the addition of some noise to cover up the obvious moves.

Such a combination is a good way to reduce a pattern unobtrusively. Ask yourself, which version does the best job of reducing the pattern while retaining realism? The answer is, the par version. Why? Because the five people who made its parents used five different approaches to attack the pattern. All of them left some kind of artifacts, but they’re no longer visible because each is only seen at 20% strength in the par version. 

706 This person was particularly careful with the darker area of the image. To get more weight into the fleshtone, he multiplied through a blurred layer mask based on the red channel. This is the correct choice, since the red is lightest in the faces. Multiplication produces a softer look than direct blends into the red channel and gets pleasing faces here.

For all that effort, this is a junior version of #703. I’ve suggested that everyone, when they think they’re done with an image and have returned it to RGB, always test Auto Tone. Hopefully this will rarely make a difference; if it does you’re probably made a mistake. And that’s the case here. There’s no point in being fastidious about protecting highlights here. Nobody is going to care if the whites in the background blow out, or even the white stripes on the tallest girl’s dress. Applying Auto Tone to #706 increases snap considerably and makes it competitive to the par.

707 This person used H-K and two Hammer actions to get decent shape in the people seated on the wall, although the overall appears too brown. One original idea: he vignetted these people, artificially darkening the folks at left of the image and making it appear that a spotlight is shining on the main subjects. In a tropical setting with the sun almost directly overhead this doesn’t make a lot of sense to me, but it does in many other image categories.

708 Accurate color, but very washed out. I’ve already discussed three solutions, which can all work in tandem where needed: 1) blending the green into the red, Darken mode, on a Luminosity layer; 2) multiplying the image through a blurred red layer mask; 3) The H-K action.

709 The person’s listing of steps doesn’t explain what’s wrong here: the subjects, and particularly me, look like they’ve been cut out of the picture and pasted back in. I surmise that the background wall was selected and then lightened. Also, the eyes needed to be excluded from the attempts to remove the pattern.

710 Chosen for the par version. Robin Mark D’Rozario posted a full description of his procedure. Basically he and Gerald were striving for the same yellowish look but this is more successful than Gerald’s #701. I still think this is too washed out. Also, the pattern reduction posterized my hair.

711 Strong purple cast, easily measurable in the wall and in the people’s hair. Remember, someone with black hair should be 0a0b or close, someone with dark brown hair should have AB equal and positive, anyone with lighter hair the B should definitely be higher. Having the A higher than the B is rare and should be confined to dark-haired people. Here, the woman at bottom left, whose hair is relatively light has an A value 14 points higher than the B.

712 See #711, way too magenta.

713 Lile #702, this is quite a pretty presentation but no real effort was made to eliminate pattern. The person reports that the entire correction took only a few minutes. He got the color essentially right, then used the green>red Darken blend as others did, plus the Bigger Hammer action at an opacity of 60%.

714 The chilliness of this one, submitted by John Furnes, was discussed in another post.

715 Chosen for the par version  This one, by John Lund, like #714 was discussed in the main thread. The yellow color is nice and the weight is about right, but John admits being one of the people who took blurring out the pattern too far and got fuzzy reproduction.

716 Rather chilly, face badly blurred. Submitted by mistake in Prophoto RGB. The strong wall is the result of running H-K.

717 Attractive color and weight. My face is a little flat, although this guy did make one version with the green>red Darken move. He then blended it into another. Not clear what was done to reduce the pattern but it seems acceptable.

718 This is basically pretty good but hurt by two factors: first, the guy produced some sort of vignette to try to lighten up the area surrounding the parakeet. This had the unpleasant effect of seeming to put my face into shadow.

More important, my face was damaged by a misapplication of the Dust & Scratches filter, which this guy used on the L channel having blurred the A and B. He chose a Radius of 3 and a Threshold of 0. When Threshold is 0 the Dust & Scratches and Median filters are identical. Putting a non-zero Threshold value in D&S tells the filter that any differences less than the threshold are natural detail that should be left alone, but anything greater is garbage that should be blurred out. So, the higher you can set the Threshold, the more natural-looking will be the result.

It follows that anyone using D&S should be using one setting for my hair and another for my skin, because the noise is extremely pronounced in the hair and not so much in the skin. It happens that in #420, I used D&S with a Radius of 3 also, but I put a Threshold of 30 or so in the hair and 15 or so in the skin. Compare the two faces and hair, and you’ll see that #718 looks plastic and posterized.

719 Chosen for the par version. Colorful but believable. Extreme, though acceptable, contrast added to my face. The FFT filter used to reduce the pattern created the sensation of noise in the face.

720 Chosen for the par version   This one is mine and I don’t object to its strong coloring. This is the tropics, after all. H-K was helpful in keeping it somewhat subdued. Anyone who says it’s too colorful should remember that seeing it next to 27 less colorful versions tends to bias the judgment.

721 The purplish cast in this one, by James Gray, was discussed in the main thread.

722 If you find the fireworks of #719 and #722 too much, this is a reasonable alternative, which good contrast and probably the best rendition of the sloths. The color measures as slightly purple, certainly not in the league of ##711, 712, and 721, but if there’s going to be a cast I’d vote for something sunnier, #715 for example. In the main thread, I suggested that although a certain amount of pattern remains, I do not find it objectionable, and certainly prefer it to many of the overly blurry ones described above. The workflow as described is complicated with a lot of individual selections, but handled carefully. For example, he painted a new fleshtone into my face in Soft Light mode. Effective and inconspicuous. 

723 Good color but washed out and blurry. See comments to #708 for solutions.

724 Too light, colors not pronounced but acceptably conservative. Pattern still in face although reduced. The individual reports unsatisfactory results with FFT.

725 Yellowish and somewhat washed out. At first, I thought this is very similar to Gerald’s #701, but when I compared the two this one is better. The apparent reason should be interesting to fans of MMM.

The original MMM as incorporated in the PPW panel alters hue in the entire image to create variation in the user-defined area. Usually that works well, sometimes not so well. Effective with v.5 of the panel we incorporate an action that Gerald wrote and calls MMM Finetuned. It tries to restrict the effect to things that are fairly close to the user-defined area.

First, let’s admit that MMM is not particularly effective in this image. My face is too small to really appreciate the effect. The clothing is all of different colors and we don’t need MMM to help us break them apart. PPW does have an advantage on this image, but it’s the channel blending, the hammers, and H-K, not so much MMM.

Here, both people defined the faces anyway to give MMM a try. #701 uses Gerald’s version and #725 the standard. Using standard MMM everything gets pushed away from red, which is another way of saying, toward green. And I can read a green cast in the stone wall. This would ordinarily be a bad idea but in #725 it seems to work, it makes the faces, which are not strongly colored, stand out better.

726 Chosen for the par version and apparently well liked by the group in spite of rather subdued color. The definition is excellent. By request, I post what the individual supplied as his steps:

Hi Dan,
Hope you're doing well. Here's my attempt at a memorable snapshot. I used the Photoline application for my adjustments.
1) Set aims (highlights/shadows/gray balance)
2) Round trip through FFT filter to mitigate repeating embossed pattern on print
3) USM ~ 15pix @ 80% (maybe too aggressive) on L channel
4) H&S to increase saturation and adjust hues
5) Lasso Dan's face/neck - H&S to adjust red balance, add density
Also several blending steps - if the color got out of hand, but I liked the lumi, then I’d blend that into a previous versions color. Likewise, if the lumi went south, I’d blend the good color into a previous versions lumi.
All told (many short sessions), took about an hour.

727. Nice color but somewhat washed out, and quite blurry. Going over the steps it appears that his favored tool against the pattern was the Median filter. As I pointed out in the comment to #718, Median is just Dust & Scratches with a zero Threshold, so there’s never any reason to prefer it to D&S. So, this skin has the same plastic, posterized look as #718, and the suggestions there apply to this one too.

728 The par version.

Dan Margulis

Dan -
Thanks for taking the time to comment on all the entries. Mine is #706, and your reminder to double-check everything with a quick Auto Tone is valuable.

Auto Tone is a nice yes-but-could-it-be-better step: I’ll keep it in my bag o’ tricks. As we all know, staring at an image too long in one sitting tends to change the way we perceive and judge color and tone in that moment. I often return to an image the next morning and wonder how I possibly thought that was a good choice! 

Rick McCleary

Dan Margulis

On Jul 1, 2020, at 3:01 PM, Dan Margulis via <dmargulis@...> wrote:

Here are my notes on this set.

We had more trouble with pattern reduction in this exercise than might be hoped for. Part of the reason was that many people were using the Dust & Scratches filter for this purpose, but their notes indicated that at least two people, and likely more, were misusing it.

If you think you might be one of them, it’s probably worth reviewing the following discussion of how Dust & Scratches works in various settings, and how it compares to Surface Blur.

Dan Margulis